the risks of aI and how blockchain can solve them2025ContentsContentsDisclaimerResearch Team&Acknowl.
2025-10-30
40页
5星级
Quantum Technologies:Key Opportunities for Advanced Manufacturing and Supply ChainsW H I T E P A P E.
2025-10-30
31页
5星级
TEST ANALYSIS OF WBBA&GSMA VOICE OVER WI-FIMEMBERSMEMBERSMEMBERSOBSERVERSPARTICIPATING INSTITUTIONS .
2025-10-30
44页
5星级
BeyondKPMG.Make the Difference.October gs1hk.org Age of AIRetail in the 2025 KPMG,a Hong Kong SAR pa.
2025-10-29
53页
5星级
November 11 12 2025Sheraton Austin Georgetown,TexasTHE GLOBAL STATE OF GENERATIVE AI IN ENTERPRISE INDUSTRY REPORT 2026Visit SiteVisit SiteRegisterState of Generative AI in the enterprise 2 2Sam LehmannEvent DirectorGenerative AI WeekWhen we released our first state of the industry report in 2023,enterprises were going through a wave of experimentation,trying to identify Gen AI transformative use cases across workflows.2024s report showed an explosion in pilot studies and proof of concepts,with enterprises seeking to define governance policies,infrastructure requirements and value creation.This years report shows how fast the landscape is evolving,as enterprises moving from pilot into full-scale production,effectively deploying and scaling Generative AI initiatives to deliver tangible business value.In this report,we seek to highlight the key forces shaping Generative AI adoption:Generative AI in Core Industries:How sector-specific use cases are evolving and whats workingAI Industry Trends:Where the technology is heading and whats driving the next wave of innovationAI in the Enterprise:What best-in-class operationalisation looks like-from architecture to governanceGen AI Investments:Where capital is flowing and how its reshaping the competitive landscapeGen AI Infrastructure:How leaders are building scalable,flexible,and cost-effective platforms for AI deploymentAs we convene at Generative AI Week,this report is designed to ground our conversations in real data,real strategy,and real outcomes.Its not just a snapshot of where we are today-its a guide to whats next for enterprise leaders seeking to implement Generative AI across E2E workflows.ForewordForewordTable of Contents and ForewordList of TablesList of FiguresEnterprise Market&Technology LandscapeGenAI in Core IndustriesGenAI in Financial ServicesGenAI in Creative IndustriesGenAI in RetailGenAI in ManufacturingGenAI in HealthcareGenAI in EducationGenAI in TransportationAI Industry TrendsAI Infrastructure&ArchitectureAgentic AIAI Governance Risk,Compliance,Responsible AIGenAI in Enterprise:Case StudiesGenAI TechnologyGenAI and InvestmentsGenAI Infrastructure DevelopmentValue Creation through GenAIVendor LandscapeAppendixBibliography3 33 34 4101010111417202425373739394343454547474848505057572828282935Contents2 2 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise TABLE 1Business functions where enterprises are using GenAI by industry(%)TABLE 2GenAI use cases across the creative industries value chainTABLE 3Impact of GenAI on the retail value chainTABLE 4GenAI applications across the manufacturing value chainTABLE 5Categorization of GenAI models in manufacturingTABLE 5Agentic AI vs GenAI vs Traditional AITABLE 6Agentic AI vs GenAI vs Traditional AITABLE 7Notable RAI policymaking milestonesTABLE 8Significant model and dataset releasesTABLE 9Leading GenAI models and specificationsTABLE 10Illustrative capabilities of GenAI platforms from select frontier labsTABLE 11Top private equity deals in Gen AI Q1 2025TABLE 12Top private equity deals in Gen AI Q1 2025TABLE 13Significant AI model and dataset releases,2024 onwardsTABLE 14Leading vendors:GenAIFIGURE 1GenAI impact on business revenuesFIGURE 2GenAI implementation statusFIGURE 3Global enterprise GenAI market by segments in US$billions,2025-2030FIGURE 4Global enterprise GenAI market by region in%,2025-2030FIGURE 5Enterprise GenAI:Market share of LLMs in 2024 in%FIGURE 6Gen AI opportunity by function in US$billion:BankingFIGURE 7Air concept shoe by GenAIFIGURE 8Potential with GenAI in educationFIGURE 9GenAI adoption and impact in transportationFIGURE 10GenAI infrastructure funding in 2024FIGURE 11Global Agentic AI market size in US$billions,2025-2030FIGURE 12Evolution to multimodal GenAI agentsFIGURE 13GenAI vs Agentic AI approach to task completionFIGURE 14Comparative scoring of leading Agentic AI solutionsFIGURE 15Investment in responsible AI by company revenue,2024FIGURE 16Leading GenAI AI chatbots market share and user growth in the U.S.,April 2025FIGURE 17GenAI spending vs economic potential of the industryFIGURE 18VC investments in GenAI,2014-2024,US$Millions6 67 7TablesFigures1313161619192020202030303636404041414242434344448 88 89 99 91515272724242929313134343535424243434444484833331010343449493 3 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise According to a 2025 U.S.-focused study by McKinsey,as many as 71%of the organizations use GenAI in at least one business function,up from 65%in early 2024.Therefore,it is no surprise that global GenAI spending in the enterprise is estimated to grow from US$4.0 billion in 2025 to US$19.2 billion in 2030 at a CAGR of 36.8%.While 2024 marked the year that GenAI became a strategic imperative for the enterprise,as companies scaled and learned from their pilots,2025 has begun to witness efforts to deliver a tangible return on investment(ROI)by deploying GenAI at scale.However,senior decision makers are not expected to demand tangible value and financial results immediately and are operating with a medium to long-term timeline.After all,despite GenAIs meteoric rise over the last two years,it is still very much in its nascent stages of development and usage,as is evident from the fact that 60%of enterprise GenAI investments today come from innovation budgets.However,with 40%of the spending sourced from more permanent budgets,58%of which is redirected from existing allocations,businesses are demonstrating a growing commitment to AI transformation.Another reason GenAI will take long to deliver tangible value is that companies need to deploy their limited resources across various competing transformational priorities and a complex and ever-changing regulatory landscape.Another point to consider is that not all enterprise GenAI investments will be fruitful.In fact,according to estimates by Gartner,at least 30%of GenAI projects will be abandoned after proof of concept by the end of 2025 due to poor data quality,inadequate risk controls,escalating costs and power requirements,or unclear business value.In fact,according to Carly Davenport,VP at Goldman Sachs,the U.S.will have to spend over US$7 billion annually in capital investment to facilitate GenAI-related new power generation alone.Additionally,they will also need to build the supporting infrastructure,such as the transmission wires that transport electricity over long distances and distribution cables that carry electricity to homes,making the overall investment much higher.Enterprise Market&Technology LandscapeWhile 2024 marked the year that GenAI became a strategic imperative for the enterprise,as companies scaled and learned from their pilots,2025 has begun to witness efforts to deliver a tangible return on investment(ROI)by deploying GenAI at scale.However,senior decision makers are not expected to demand tangible value and financial results immediately,and are operating with a medium to long-term timeline.4 4 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Even though investments in foundation models still dominate enterprise GenAI spending,the application layer is now growing faster.The top three areas of GenAI application spending are mentioned below.Code copilotsThe intersection of AI and coding has become one of the hottest areas in the technology world regarding VC investments.AI coding tools can automate various routine development tasks such as code generation,testing,and debugging,which has proven to be particularly useful given the huge global demand for software and the shortage of skilled developers.GitHub Copilots rapid rise to a US$300 million revenue run rate validates this trajectory,while emerging tools like Codeium and Cursor are also growing fast.Beyond general coding assistants,enterprises are also investing in task-specific copilots like Harness AI DevOps Engineer and QA Assistant for pipeline generation and test automation,along with AI agents like All Hands that can perform more end-to-end software development.Support chatbotsAccording to the Menlo Ventures study,support chatbots attracted 31%of enterprise adoption in 2024.A good example is global bank ING,which has managed to resolve around 45%of its 85,000 weekly customer interactions in the Netherlands alone through chatbots.Aisera,Decagon,and Sierra are examples of agents that interact directly with end customers,while Observe AI supports contact center agents with real-time guidance during calls.Enterprise search&retrieval and data extraction&transformationenterprises are investing significantly in these solutions to unlock and harness the knowledge often hidden within data silos across organizations.Good examples are solutions such as Glean and Sana that connect to emails,messengers,and document stores to enable unified semantic search across disparate systems and deliver AI-powered knowledge management.The intersection of AI and coding has become one of the hottest areas in the technology world regarding VC investments.AI coding tools can automate various routine development tasks such as code generation,testing,and debugging,which has proven to be particularly useful given the huge global demand for software and the shortage of skilled developers.1 1 2 2 3 3 5 5 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise INDUSTRY BUSINESS FUNCTIONSMarketing and salesTable 1:Business functions where enterprises are using GenAI by industry(%)TechnologyProfessional ServicesAdvanced IndustriesMedia and TelecomConsumer Goods and RetailFinancial ServicesHealthcare,Pharma,MedicalEnergy and MaterialsOverallProduct and/or service developmentITService operationsKnowledge managementSoftware engineeringHuman resourcesRisk,legal,and complianceStrategy and corporate financeSupply chain/inventory managementManufacturingUsing gen AI in at least 1 function55393130263616121410588494116233491791443804839262417171362115137945262237263022610337946212013128811714868402524261620112174065292230142413756256333172613138169567594228232221181311117571Note:Global survey conducted between July 16-31,2024,with 1,491 participants at all levels of the organizationSource:McKinseyMarket sizeThe global enterprise GenAI market is estimated to grow from US$4.0 billion in 2025 to US$19.2 billion in 2030 at a CAGR of 36.8%.One of the main reasons for the technologys growing popularity across the enterprise is the public availability of advanced and breakthrough GenAI tools such as ChatGPT,Googles Gemini,and Microsoft Copilot,which have made professionals comfortable with the potential use cases for more industry-centric use.Even though there is consistent adoption across industries,some of them,such as information technology(IT),cybersecurity,operations,marketing,and customer service,are more mature than others.Moreover,enterprises that reported higher ROI for their most scaled initiatives are broadly further along in their GenAI journeys.6 6 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise The global enterprise GenAI market is estimated to grow from US$4.0 billion in 2025 to US$19.2 billion in 2030 at a CAGR of 36.8%.One of the main reasons for the technologys growth is the public availability of advanced and breakthrough GenAI tools such as ChatGPT,Googles Gemini,and Microsoft Copilot,which have made professionals comfortable with the potential use cases for more industry-centric use.The software segment is expected to account for the largest 67%share in 2025,with services accounting for the rest.The emergence and expected meteoric rise of AI agents is the primary driver of the software segment over the short to medium term,as the technology gains interest as a breakthrough innovation with the potential to unlock the full potential of GenAI.However,it should be noted that agentic AI cannot be considered a silver bullet,and all the broader challenges currently facing GenAI still apply.Figure 1:GenAI impact on business revenues Note:Global survey conducted between Feb 22-Mar 6(H1 2024)and Jul 16-31(H2 2024).A question was asked of those who said their organizations regularly use GenAI in a given function.Source:McKinsey&CompanyIncrease by 10%Increase by 610%Increase by 5%First half of 20241Strategy and corporate finance7355Supply chain and inventory management18307Marketing and sales12343Service operations13297Software engineering9304Product or service development823%Second half of 2024111247191532824341814311213311215257 7 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Figure 2:GenAI implementation statusNote:The ISG GenAI Use Case Study,conducted in August 2024,surveyed 2,000 companies globallySource:ISG 2024 MarketLens AI StudyLive/pilotMoving towards full implementationFully implemented,Evaluation stageNot live(Testing phase)43%8%7%Figure 3:Global enterprise GenAI market by segments in US$billions,2025-2030Source:AgileIntelSoftwareServices203019.22025202620277.620289.9202913.98 8 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise North America is expected to dominate the global market,accounting for approximately 41%of the market share by 2025.Meanwhile,pthe Asia Pacific is projected to be the fastest-growing region from 2025 to 2030,with significant contributions from China,Japan,South Korea,and India,driven by substantial government initiatives.OpenAI dominates the market with a share of 32%,followed by Anthropic(25%),Meta(15%),Google(13%),and Mistral AI(5%).Figure 4:Global enterprise GenAI market by region in%,2025-2030Source:AgileIntelNorth AmericaEuropeAsia PacificLatin AmericaMiddle EastAfrica40.53.7.0%5.2%1.6%Figure 5:Enterprise GenAI:Market share of LLMs in 2024 in%Note:Metas Llama 3 and Mistral are open-source LLMsAgileIntelOpenAI32%Mistral5%Anthropic5%Cohere3%Meta15%Internal model3%Google13%Others4 25203040.62.8.4%5.4%1.8%9 9 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise GenAI in Financial Services The global financial services industry continues to operate in volatile macroeconomic conditions characterized by sudden interest rate hikes and heightened trade tensions.While European and Indian banks are reaping the rewards of rising interest rates,North American banks face a mixed bag of results due to more polarized outcomes.On the other hand,just when Japanese banks had begun to show signs of recovery,U.S.tariff fears resulted in the countrys banking index plunging over 20%in the week ending April 4,2025.Amidst such uncertainty,only the banks that adapt will thrive while the others risk being left further behind.One key adaptation strategy employed by the global financial services industry is the integration of GenAI,which has become a core enabler of banking transformation.The technology has the potential to not only enable operational transformation and reinvent business models but also save costs,generate higher revenues,and address risk and compliance requirements.Moreover,as the industry becomes more digitized,GenAI offers opportunities to automate complex processes,deliver customized customer experiences,and strengthen security measures,thereby allowing them to compete with nimbler digital-first competitors.This is especially important in todays volatile macroeconomic environment,which has placed significant pressure on global financial organizations to deliver adequate returns to stakeholders.According to McKinsey estimates GenAI could add between US$200 billion to US$340 billion to the global banking sector annually.GenAI in Core IndustriesSource:KPMG,February 2025Figure 6:Gen AI opportunity by function in US$billion:BankingFinanceHRITSerivce and data analyticsCyberRiskOps&Supply ChainMarketingSalesOther3252093554116207010 10 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Achieving ROIEven though several financial services companies have already successfully implemented GenAI in their operations and started realizing efficiencies,only a few have reported achieving revenue growth from their GenAI investments.Therefore,they now face significant pressure from shareholders to show immediate ROI on their investments.However,despite these pressures and the uncertainties created by the rapid evolution of AI technologies,global financial institutions are poised to increase their GenAI budgets over the short to medium term.In fact,according to a 2025 study by BCG,one in three banks plan to spend over US$25 million on ramping up their GenAI capabilities in 2025.However,there is a significant shift in how GenAI is being deployed across the banking industry as banks and other organizations shift from broad experimentation to a strategic enterprise approach that prioritizes targeted applications,especially at the interface between institutions and customers.GenAI-powered tools now support autonomous chat agents that transcend predefined scripts,real-time loan approvals,and automated processing of submitted documentation.Interestingly,enterprises view the potential value of GenAI in the financial services industry not only as a downstream application but as a tool that complements other machine learning(ML)models and applications.Therefore,they are integrating GenAI not as stand-alone silo models but as a part of a network of models and technologies including robotic process automation(RPA)and autonomous agentic AI solutions.Here,the insights and outputs from one are used to inform the function and direction of another.This approach has already started to deliver results in the form of 24/7 virtual advisors,providing customized financial guidance,automating routine transactions,and proactively managing customer needs based on real-time data and predictive insights.Additionally,back-office processes,such as fraud detection,compliance monitoring,and risk assessment,are getting streamlined by analyzing vast amounts of data with enhanced speed and precision.GenAI in Creative Industries The creative industry has historically relied heavily on human intuition,emotion,and originality,protecting it from disruption by AI and related technologies.However,GenAI has opened up many opportunities,with the sector now ripe for an imminent disruptive impact.Even though several financial services companies have already successfully implemented GenAI and started realizing efficiencies,only a few have reported achieving revenue growth from their GenAI investments.Therefore,they now face pressure from shareholders to show immediate ROI on their investments,with a 2025 study by KPMG pegging this at around 70%.Among GenAIs most promising applications in the creative industries is the use of conversational interfaces to create novel content or translate existing ones.For example,the technology can be used to generate videos or podcasts from articles and blog posts,or to generate variations of a script or storyboard,enabling creators to explore options faster.11 11 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise This is mainly due to the technologys ability to not only automate repetitive tasks such as resizing images,removing backgrounds,and generating design variations,but also provide a new palette for creative individuals to experiment with.This includes generating personalized content,pictures,and videos that are virtually indistinguishable from those made by humans,enhancing operational efficiency,and enabling companies to quickly adapt to evolving trends.In fact,according to a June 2024 article by BCG,GenAI can now create high-quality content at near-zero marginal cost that allows companies to deliver on the promise of personalization at scale.Another study by the World Economic Forum(WEF)showed that GenAI tools can save creative professionals up to 11 hours per week on tasks such as brainstorming,prototyping,and refining content.These benefits empower more people,including those without deep technical or artistic skills,to join the creators board.Among GenAIs most promising applications in the creative industries is the use of conversational interfaces to create novel content or translate existing ones.For example,the technology can be used to generate videos or podcasts from articles and blog posts,or to generate variations of a script or storyboard,enabling creators to explore options faster.Text-to-video GenAI models such as OpenAIs Sora have spurred a tectonic shift in the advertising industry,with brands and agencies innovating at a rapid pace to leverage AI-generated video content in their advertising.LLMs,Generative Adversarial Networks(GANs),Deep Reinforcement Learning(DRL),and Multi-Modal GenAI are the four main GenAI technologies that underpin much of this disruption.LLMs can generate human-quality content,such as poems or scripts,much faster than people can,and also translate languages.GANs go a step further by pitting two neural networks against each other,with one creating new content and the other evaluating its authenticity.They can also produce advanced imagery,ranging from photorealistic landscapes to abstract compositions.DRL employs a reward-based,trial-and-error system through which AI agents can create content that aligns with specific aesthetic preferences or user behavior patterns.Multi-modal AI works by learning patterns and the association between text descriptions and corresponding images,videos,or audio recordings.The impact of GenAI in creative industries is already visible.A good example is Adobe integrating related capabilities throughout its Creative Cloud suite,with tools like Generative Fill and Text to Image,which are changing how designers work.According to Scott Belsky,Adobes former Chief Product Officer,the company is aiming to have a language user interface for all of its applications over the short to medium term.Another is graphic design software company Canvas Magic Studio,which has democratized design by making sophisticated AI tools accessible to non-designers.In 2025,the use cases that are expected to gain traction include:Applications such as U.S.-based Runway AIs text-to-video tool and Cinelytics analytics and predictive film intelligence platform are designed to plug into production workflows,enabling studios and filmmakers to streamline production tasks and make informed business decisions.1 112 12 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Tools such as Pencil AI,which is built on OpenAIs GPT LLMs,can create high-quality,low-cost ads quickly,with predictive analytics to test performance.ChatGPT also provides analytical capabilities,allowing industry players to create audience archetypes to test new TV programs.From a post-production perspective,AI applications that provide dubbing and subtitling solutions are expected to witness increased usage.Platforms like Speechify,ElevenLabs,and Panjaya.ai simplify and expedite the process of dubbing audio and creating closed captioning.This enables distribution companies to generate incremental revenues in areas where localization costs have historically been expensive.GenAI-based music generation tools such as MuseNet,Magenta Studio,and Musicfy that can assist in composing music by learning complex musical patterns,predicting the next word or music note in a sequence,and mixing specified instruments.They can also change one type of sound into another,such as from whistling to the violin or from the flute to the saxophone.This capability is beneficial for artists who may not be proficient in playing all the instruments they wish to incorporate,saving both time and costs.This space has advanced rapidly due to unsupervised learning on large datasets and the use of transformers.Image generation tools such as Stable Diffusion,Midjourney,DALLE,and Ideogram,based on diffusion models(DMs),are fast gaining traction.These open-source tools are developed with the Multimodal Diffusion Transformer(MM-DiT)architecture,which is beneficial for both text and image.Commercial AdoptionBusiness OperationsCommercial StrategyPost-productionProductionPre-productionSource:Alix PartnersTable 2:GenAI use cases across the creative industries value chainLow Customer service chatbots Content moderation Personalized content discovery Dynamic and personalized advertising AI dubbing for content localization Content moderation Media content for publishing(text&image)Audio content generation Concept development for marketing campaigns Market analysis Market testingMediumHigh AI-integrated VFX workflows(storyboarding,motion capture)Movie predictive analysis Script analysis Game prototyping Script writing Cybersecurity and protection Streaming optimization Conversation summarization tool Voice cloning Creating realistic sound effects for film,TV,or games Video editing process automation News article generation Music composition AI-based virtual reality experience AI rendering Budget management Conversation summarization tool Coloring and grading Visual effects(VFX)workflow AI news broadcaster Autocompleting code to assist in-game programming AI game NPCs2 23 34 45 513 13 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise GenAI in RetailAccording to an April 2024 study by McKinsey involving many Fortune 500 retail executives,as many as 82%of the respondents said that even though they were still in the piloting and testing phase,the technology had big potential,mainly in augmenting their internal value chains.In 2025,most of the pilots and proofs of concept are expected to assume a larger scale and start delivering ROI,especially in terms of faster,real-time actionable insights in minutes or days,compared with weeks or months earlier.The technology,especially conversational AI,is democratizing data analysis,allowing non-technical users to derive meaningful insights without the need for specialized skills.This not only speeds up internal decision-making but also enables more flexible and innovative use of information across the retail industry.GenAI is also expected to impact other areas of the retail value chain with automation of routine tasks such as employee scheduling,predictive maintenance,customer inquiries,and onboarding new employees,witnessing maximum disruption.According to McKinsey estimates,GenAI is poised to unlock between US$400 billion to US$600 billion in economic value for retailers and resolve billions of dollars in inefficiencies.It is also expected to reduce forecasting errors by up to 50%,helping retailers keep up with consumer trends.Therefore,it is no surprise that 45%of global retail marketing leaders plan to invest in GenAI over the next 12-24 months,according to a recent study by Deloitte.Another study by research and advisory company IHL Group found that GenAI is poised to increase retail sales by 51%and gross margins by 20tween 2023-2029,while reducing selling and administrative(S&A)costs by 29%The main challenge facing the industry in terms of GenAI deployment is that most of the companies are heavily reliant on existing,general-purpose tools.A late 2024 report by PYMNTS Intelligence involving over 500 C-suite employees in the U.S.retail industry found that 61%of them are using just existing baseline models,limiting their ability to achieve more transformative ROI.Comparatively,sectors such as information and manufacturing were ahead in developing proprietary solutions,with 70%and 69%doing so,respectively.Key use case opportunitiesRetail Media:presents a high-margin opportunity for retailers who are increasingly selling their data to brands that can then leverage it to reach consumers closer to the point of purchase.Advances in GenAI are expected to augment retail media by automating ad campaign creation and optimization and helping brands enhance their return on ad spend(RoAS).It is also likely to improve both self-serve and programmatic ad-buying infrastructure due to its ability to process millions of data points within seconds,helping media buyers select the optimal ad format,including the time and location the ad will air.According to a January 2025 study by Coresight Research,the U.S.retail media market is expected to reach US$67.8 million by the end of 2025,ultimately increasing to US$106.4 billion in 2028,at a CAGR of 16%.14 14 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Product development:GenAI models offer brands various ways in which they can improve their creative processes in terms of new product development.While multimodal models,such as Midjourney,have offered image-generation features for some time,new GenAI applications allow creative professionals to deploy these models without learning how to design prompts and interact with them.Additionally,applications such as Digital Wave Technologys Maestro allow brands to generate creative new product ideas that are more consistent with the brand story and avoid hallucination and toxicity.Below is an image from NIKE,revealing the artistic possibilities of GenAI for new product ideation.Further,GenAI models can facilitate new product development by mining social media posts for major or emerging customer trends or analyzing product reviews,which can then be input into image-generation applications for new product ideas.2025 is expected to witness the availability of applications that can manage and control multiple GenAI models,which will democratize the use of image-generation technology,making it accessible to a wider base of non-technical users.Voice commerce:2025 is expected to witness the expansion of GenAI-powered voice-based shopping or V-Commerce,allowing users to complete purchases,receive customized recommendations,and manage orders using voice commands.A good example is Apple Intelligence,which has integrated advanced natural language capabilities in Siri to offer highly customized shopping recommendations and even predict future purchases.Another disruptive example is SoundHound AI,which is integrated in vehicles,allowing drivers and passengers to order takeout for pickup directly from the cars infotainment system through voice commands.Examples of retailers using GenAIAmazon:has developed an AI virtual assistant called Rufus that is trained on the companys product catalog and customer reviews,among other resources.The application leverages Amazon Web Services(AWS)chips Trainium and Inferentia,and a custom-built LLM that allows it to answer product-related questions and compare products,in a personalized setting.CarMax:a U.S.-based car retailer,was one of the first in the industry to start using GenAI and has since evolved the technologys usage to create detailed car comparisons with specifications,features,benefits,and customer reviews.Its internal tool,called Rhode,simplifies access to company knowledge for associates,while Skye augments customer experience during vehicle transactions.The North Face:has deployed IBMs Watson-powered GenAI model to offer a conversational shopping assistant on its online shopping platform.The AI assistant asks customers questions about their preferences,planned activities,and intended usage for outdoor gear,and then delivers product recommendations based on the responses.Figure 7:Air concept shoe by GenAI Note:Air concept for tennis player Zheng QinwenSource:NIKE15 15 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise eBay:The companys GenAI-powered shopping assistant,ShopBot,helps customers navigate through over a billion listings using text,voice,or even by sharing a photo to indicate what theyre searching for.The bot can also initiate further conversations to enhance its understanding of the customers requirements,thereby allowing it to offer tailored suggestions.Shopify:has launched a GenAI tool called Magic that uses automatic text generation to create automated content such as product descriptions,email subject lines,and headers for an online store.It also allows merchants to modify photo backgrounds to suit their branding,without needing expertise in complex software like Photoshop.Retail value chainAfter generative AIBefore generative AISource:McKinsey&CompanyTable 3:Impact of GenAI on the retail value chainProcurement GenAI chatbots handle initial rounds of supplier negotiations GenAI-powered briefs and summaries of supplier terms assist procurement associates in closing deals.Manual handling of supplier negotiations(including end-to-end contract creation),often resulting in overlooked details Tedious supplier assessments based on limited data,leading to suboptimal choicesDistribution Initial communication and email messages to third-party logistics handled by Gen AI chatbots Returns management process,along with a response to distribution disruption,supported by Ge AI Individuals handling communication with third-party logistics providers Delayed response to distribution disruptions due to the complexity of supply chain operationsIn-store operations People use GenAI-powered assistants for instant voice access to information Information searches,such as price,in-store location,and stock level handled manually by associates,leading to delayed customer serviceE-commerce Automated generation of e-commerce content(eg,product profiles,descriptions)within a few minutes E-commerce customer experience personalized spontaneously by automated front-end development techniques Hundreds of hours spent on the generation of e-commerce content Manual rule-based website personalization,consuming employees resourcesMarketing Unlimited insights extracted from different unstructured sources(eg,product reviews)Fully personalized marketing materials generated with increased efficiency for every customer One-size-fits-all marketing approach due to limited customer insights derived from structured data Creation of marketing materials through a lengthy,iterative processBack office The next-generation“white collar”leantransferring administrative processes of support functions to GenAI-powered chatbots and interfaces,such as development copilots,HR/financial copilots.Time-consuming administrative processes,such as HR and payroll,are prone to errors and inefficienciesAccording to McKinsey estimates,GenAI is poised to unlock between US$400 billion to US$600 billion in economic value for retailers and resolve billions of dollars in inefficiencies.It is also expected to reduce forecasting errors by up to 50%,helping retailers keep up with consumer trends.16 16 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise GenAI in ManufacturingOver the past couple of years,GenAI has transitioned from a futuristic concept to a tangible transformative force,shaping the manufacturing landscape in previously unimaginable ways.The technology now allows manufacturers to automate and enhance factory activities by supporting functions such as programming and machine maintenance(including predictive maintenance),autonomous factory management,intelligent quality control,smart supplier contract management,and product R&D.A good example is German manufacturer Bosch which is using GenAI to create a comprehensive dataset of synthetic product defect images to train its AI system for optimal quality control.According to a 2025 study by Deloitte titled Future of Manufacturing involving 600 manufacturers globally,as many as 87%reported that they had initiated a GenAI pilot already,while 24%indicated that they had adopted GenAI use cases in at least one of their facilities.Further,50%of the respondents said that GenAI solutions ranked among the top-priority solutions for their organizations over the next 24 months,higher than other highly sought-after technologies such as digital twins,the omniverse,and the metaverse.Another 2025 study by technology company NTT DATA involving over 500 manufacturing leaders and decision makers in 34 countries,a staggering 95%of them said that GenAI was already directly improving efficiency and bottom-line performance.Interestingly,94%expect the integration of IoT data into GenAI models to significantly improve the accuracy and relevance of AI-generated outputs.Manufacturers are also using GenAI to personalize operations by training LLMs on smaller datasets from their internal industrial IoT(IIoT)devices,instead of the conventional large datasets.This enables seamless information exchange between legacy machines and equipment not using open-source AI tools and GenAI systems.Additionally,these smaller language models can be fine-tuned to operate closer to the edge(end-user),where latency and security are important to IIoT solutions.GenAI-powered robots are used in manufacturing through the use of natural language prompts that are inherent in the technology.This allows machine operators who are not necessarily trained in robotics or software code to communicate with the machines using natural language.Key benefits of using GenAI in the manufacturing industry:Faster product rollouts:GenAI tools allow manufacturers to bring products to market faster by automating and optimizing different stages of product development,including innovation,design,prototyping,and testing.Once a GenAI model has been trained on a products bill of materials,raw material usage,process parameters,internal research data,and other data(such as product patents or previous product trials),it can identify the ingredients that may be best suited for a new product,predict the products benefits,and recommend formula recipes.A good example is AstraZeneca,which is using GenAI to automate and quicken the drug development process.The technology has already helped the company reduce development lead times by 50%and the use of active pharmaceutical ingredients in experiments by 75%.Another leading pharmaceutical company is using GenAI to analyze production line bottlenecks and optimize its tablet packaging process.It has resulted in boosting production efficiency by 20%while minimizing material waste.17 17 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Digital twins:manufacturers are using GenAI algorithms to create accurate digital representations of their products,production lines,or entire factories.Real-time data is taken from sensors and other sources to improve design,test new processes,and create new products without disrupting the production process.A good example is Indian specialty chemical manufacturer Jubilant Ingrevia,which has reduced process variability by 63%by deploying digital twins to model,forecast,and manage operations in real time.New product development:GenAI tools analyze vast information on prevailing market trends,consumer preferences,and past performance of products,to give manufacturers a clearer picture of new and advanced product designs,and even discover new business models.In terms of novel designs,GenAI enables manufacturers to visualize concepts in high fidelity much earlier in the design process and get precise feedback from customers,thereby allowing them to create a previously unimagined product.McKinsey estimates that GenAI could unlock US$60 billion annually in productivity in product research and design alone.Additionally,through synthetic data augmentation,GenAI can enable accurate simulations,aligning product development with stringent requirements and customer preferences,thereby saving time and resources.Predictive maintenance:Previously,manufacturers prevented breakdowns by performing scheduled maintenance according to fixed cycles or periods.With the advent of AI and ML,they began using data from various sensors to identify patterns,predict breakdowns,and then proactively conduct maintenance.GenAI has further improved this process by automatically creating text or images that provide detailed instructions,including lists of required spare parts.This system enables maintenance personnel to spend more time on the actual tasks instead of preparing instructions,enhancing productivity,and reducing costs.Owing to its comprehensive nature,it also allows inexperienced technicians to repair or maintain equipment more effectively.Customization at scale:GenAI allows for the efficient customization of products at scale,catering to the unique preferences of individual customers without compromising efficiency.By using this technology,manufacturers can readily adjust designs and processes to meet customer demands in real time.AI-driven insights allow for the integration of unique product features on a large scale,without a significant increase in costs.As the technology evolves,the potential for personalized products will expand,optimizing design,performance,and functionality based on specific customer preferences.Industries already in the advanced stages of integrating GenAI in their manufacturing processes include consumer electronics,automotive,and fashion.18 18 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Table 4:GenAI applications across the manufacturing value chain Planning-product developmentCreate product concepts and engineering drawings to reduce R&D and prototyping times.Predict product-market fit with qualitative consumer/market data.Discover new materials by testing to define their fit and function as alternative raw materials.Planning-production planning and procurementDevelop production plans based on available materials,equipment,and resources.Pre-screen,summarize,and extract clauses of interest across contracts and assess risks.Discover new supplier profiles across sources.Automatically action ERP exception messages to achieve optimal inventory levels Production-performance,maintenance,and health and safety Create employee training videos and maintenance troubleshooting role-plays.Identify hazardous working conditions and notify key stakeholders about required measures.Write standard operating procedures and policies,and translate documents into other languages.Automate root cause analysis to identify causes of nonconformances without manual data analysis Adjust production orders in real time based on LoT,RFID,and order-tracking data.Predict exact machine failure modes and automatically develop intervention plans.Receive performance updates,priorities,and advice from Al chatbots.Supply chain-warehousing and logistics Automate route design,using routing algorithms to reduce cost and lead time Generate and verify the required documents for transportation.Provide updates on shipments and delivery times via chatbot interface.Provide an interactive virtual assistant for drivers to augment typical services provided(eg,route navigation)Optimize warehouse design to streamline order-picking routes.Improve yard management processes based on sensor and camera data.Automate materials reordering to minimize stockouts and inventory levels.GenAI has transitioned from a futuristic concept to a tangible transformative force,shaping the manufacturing landscape in previously unimaginable ways.According to a 2025 study by Deloitte titled Future of Manufacturing involving 600 manufacturers globally,as many as 87%reported that they had initiated a GenAI pilot already,while 24%indicated that they had adopted GenAI use cases in at least one of their facilities.Source:McKinsey&CompanyContent Generation Insight Generation Interaction19 19 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Table 5:Categorization of GenAI models in manufacturingGenerative AI ModelGenerative Adversarial Networks(GANs)Creation of digital twins,virtual replicas of physical assets or processes based on real-time sensor data for product design and optimizing manufacturing processes.Pros:High-quality realistic images and data augmentation,processing sequential data in parallelCons:Difficult to train,limited and repetitive outputs,difficult to find the right balance between the generator and discriminator.Application in ManufacturingVariational Autoencoders(VAEs)Prediction of equipment failures through machine learning algorithms trained on machine data.Pros:Generating data similar to training data,overcoming limitations of traditional image processing methodsCons:Less flexible than GAN,unable to tackle sequential data,difficult to control the qualityTransformer-Based Models Simulation of production scenarios,prediction of demand,defect detection,and material fracture mechanics.Pros:Processing sequential data in parallel,handling multiple data types,Powerful for diverse multimodal tasksCons:Requiring large amounts of high-quality training data,slow and computationally intensive processSource:ScienceDirectGenAI in HealthcareGenAI is rapidly transforming the healthcare industry.As many as 85%of respondents in McKinseys Q4 2024 survey of U.S.-based payers,health systems,and healthcare services and technology(HST)groups are already implementing the technology across the enterprise.Another study by Deloitte conducted towards the end of 2024 demonstrated similar results,with as many as 75%of the companies in the healthcare space already experimenting with GenAI.The widespread acceptance of the technology is driving the rapid evolution of the industry in the face of many years of plateaued growth in the areas of telemedicine and digital therapeutics.20 20 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise As GenAI matures,it is resulting in the creation of novel solutions,especially to address gaps in areas pertaining to chronic conditions such as heart failure,diabetes,and mental health.Every aspect of healthcare,ranging from personalized care to automated workflows,is expected to be disrupted at various levels by GenAI in 2025.The industry is expected to witness a greater adoption of multimodal GenAI models that can analyze and generate text,images,genomics data,and even real-time patient vitals simultaneously,compared to the single modality models that were dominant in 2024.According to a 2025 study published in the Journal of Medical Internet Research(JMIR),patients receiving care powered by GenAI attended 42%more therapy sessions and achieved a 25%higher recovery rate compared to other treatments.These findings showcase GenAIs ability to improve clinical outcomes and the overall standard of care.If 2023 was about GenAI experimentation and 2024 was about point solutions,2025 is expected to be about value delivery through end-to-end transformation.Instead of isolated GenAI tools fulfilling specific tasks like physician note-taking or scheduling,the industry is expected to witness the proliferation of integrated systems that automate entire workflows ranging from patient intakes to treatment plans.These intelligent agents will coordinate across departments,learning from each interaction to improve efficiency and outcomes.For example,in the pharma industry,key processes that will be transformed with GenAI include clinical trials,regulatory submissions,medical legal regulatory review,and omnichannel engagement.Overall,with the global healthcare industry grappling with challenges such as labor shortages,clinician burnout,declining profitability,and worsening health outcomes,GenAI offers a transformative enterprise approach to address these problems.The technology is primed to address the healthcare industrys greatest pain points by democratizing knowledge,increasing interoperability,expediting drug discovery,and enabling hyper-personalization of the care experience.Among the various areas that could witness significant disruption over the medium to long term are patient and member experience,daily administrative tasks,and clinician and clinical productivity.Despite the enthusiasm around the technologys large-scale integration in the healthcare industry,an early 2025 BCG report predicts that over 33%of ongoing GenAI programs will fail to deliver value in 2025.These failures are ultimately likely to pave the way for more sustainable and impactful transformations,driving a sharper focus on integrating GenAI into existing health care workflows.GenAI applications in the short to long term:If 2023 was about GenAI experimentation and 2024 was about point solutions,2025 is expected to be about value delivery through end-to-end transformation.Instead of isolated GenAI tools fulfilling specific tasks like physician note-taking or scheduling,the industry is expected to witness the proliferation of integrated systems that automate entire workflows.21 21 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Short term:the immediate applications of the technology are focused on the use of natural language processing(NLP)in healthcare settings,enabling functions such as ambient scribing to lessen the burden on manual clinical documentation.Other use cases are automated consumer messaging,clinical message autoreply,and document auto-generation.Medium-term:over the medium term,the technology is expected to facilitate the integration of data science within various hospital functions to extract relevant insights from data sources such as medical records,research studies,and patient-generated data,resulting in more personalized and effective treatment plans.Long term:over the long term,GenAI is expected to replace the doctor in diagnosis and prognosis.In fact,in some cases,AI and ML have already reached a 98.4curacy in certain cancer diagnoses,paving the way for quick disruption in the future.Key use cases of GenAI in the healthcare industryDrug and treatment discovery:In 2024,AI-powered drug discovery made many gains.However,in 2025,GenAI is expected to bring about rapid disruption by facilitating the design of novel drug compounds in real time.Pharmaceutical and biotech companies are increasingly using customized language models to augment their understanding of disease biology and accelerate processes to identify promising compounds.Both commercial and open GenAI models can already analyze vast biomedical data sets to suggest novel molecular structures,predict drug interactions,and design custom compounds tailored to a specific target or disease.Many of these compounds have been hard to discover through traditional methods.This mitigates formidable costs and time constraints.When used together with causal modeling approaches,the models allow companies to identify clues previously undiscovered or underrepresented in clinical data,unveiling previously ignored therapeutic opportunities.According to a recent BCG report,in 2025,this trend will further shorten discovery cycles and reveal more promising candidates to test in clinical settings.Drug development:In addition to discovery,GenAI can enhance the drug development process across all areas,such as preclinical testing,clinical study design,and regulatory submissions.For preclinical testing,GenAI models can estimate the toxicity of a drug compound by analyzing chemical structures and potential risks associated with candidate therapies.They can also forecast pharmacokinetic properties and ADME(absorption,distribution,metabolism,and excretion)properties of drug candidates,which can predict the effect of a drug on its target and related safety levels.In terms of clinical study design,GenAI increases the chances of success by identifying the most relevant patient populations,endpoints,and dosing regimens.And finally,the technology can expedite the regulatory submissions process by automating compliance checks and proactively performing checks against guidelines.Additionally,GenAI tools are poised to transform manufacturing operations by processing engineers to optimize workflows to manufacture therapeutic products,including monoclonal antibodies and cell therapies.A good example is Exscientia,a drug design and development company that uses Google Cloud GenAI capabilities to enable faster drug discovery through a Design-Make-Test-Learn(DMTL)cycle.22 22 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Quality control:GenAI is now playing a bigger role in quality control for pharmaceuticals and medical device products by standardizing the manufacturing processes and improving the detection and mitigation of related deviations.This approach to quality control will allow manufacturers to adjust processes,reduce waste,improve yield,and increase product quality.For example,an issue-resolution GenAI model trained with historical data can enable organizations to identify the effects of minor changes on product outcomes and thereby reimagine processes without extensive and often manual trial-and-error tests.Chatbots:According to an August 2024 study by health policy research company KFF,just over 16%of the adult respondents said that they use AI chatbots at least once a month to find health information or advice,rising to 25%for adults under 30 years old.As GenAI-powered chatbots evolve and improve,these consumer behavior patterns will most likely force established online health information gateways to offer their bespoke AI tools or risk losing web traffic.This will enable health providers to start realizing significant operational efficiencies and competitive advantage by using these trained chatbots to attract patients and routing them to the most appropriate sources of care,while reducing the burden on humans who staff the 24/7 triaging capabilities that they offer.Personalized care:Recent advancements in agentic AI are driving personalized treatment plans by analyzing large datasets containing patient-specific data such as genetic profiles,medical records,and live health data.This helps healthcare professionals to recommend targeted therapies such as chemotherapy,radiation,or surgery,depending on each patients unique profile.According to a study published in the ScienceDirect journal in March 2025,GenAI-powered personalized treatments improved cancer patient survival rates by 20%compared to standard care and extended progression-free periods by 15%.According to an August 2024 study by health policy research company KFF,just over 16%of the adult respondents said that they use AI chatbots at least once a month to find health information or advice,rising to 25%for adults under 30 years old.23 23 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise GenAI in EducationGenAI is transforming the education industry by disrupting traditional teaching methods,improving student support systems,and reorganizing the overall ecosystem.A late 2024 report by American education technology company Cengage Group found that as many as 49%of higher education instructors in the U.S.are already using GenAI,up from 44%in 2024 and just 24%in 2023.The technologys core capabilities,which include creating and disseminating information,make it ideal for disrupting the education space.Over the last year or so,LLMs have showcased their ability to answer questions on a range of subjects,write cogently,and even create images.Moreover,ChatGPT and similar models have proven their expertise in cracking tough examinations in fields such as law,medicine,history,and even operations management.Education technology companies and students have already started using GenAI tools such as ChatGPT,TutorAI,and the Poe app that stimulate creativity by assisting in brainstorming sessions and generating fresh ideas.Additionally,GenAI models have started assisting teachers in creating homework and assignments,explaining complex concepts to students simply,designing courses,and creating gamified learning experiences and personalized learning plans for each student.A good example is Speechify,which offers text-to-speech or speech-to-text generation capabilities that are particularly useful for students with learning disabilities such as dyslexia or ADHD.Another is Kahoot!,which uses GenAI to design games that align with curriculum goals,making learning both fun and effective.Source:Attri.aiFigure 8:Potential with GenAI in educationRadical Concepts:Teacher-less schoolsSchool-less educationsEnhanced personalization:Personalized and practicallearning experienceAugmented content creationReal-time assessmentGenAIlandscapeCourse design:Material organizationPersonalized learning pathsInteractive learning environmentsAI-assisted authoring:Content creationAssessmentAcademic research andknowledge development:Research assistanceData analysis,knowledge generation24 24 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Key use cases:Personalized adaptive learning experience:GenAI-powered intelligent learning platforms analyze various types of student data,such as historical performance,skills,and teacher feedback,to offer personalized and adaptive learning experiences.By analyzing large datasets,educators can identify knowledge gaps and provide recommendations and guidance.GenAI tool can create exercises,quizzes,and practice questions customized to each students learning needs.Additionally,through the use of GenAI tools,teachers can offer real-time assistance,progress monitoring,and adjust teaching strategies to optimize learning.Curriculum creation and design:educators are using GenAI to create course and teaching materials such as syllabi,quizzes,exercises,and concept summaries.This not only saves time through the automated generation of content,but also improves resource variety.GenAI also enables the rapid creation of e-learning capsules,micro-videos,and interactive multimedia elements,personalized to the needs of different courses.Moreover,platforms providing courses for language learning can use GenAI to correct grammar and create related exercises and questions.Virtual experiments:GenAI,together with virtual reality,is being used to make simulations and virtual environments to enable students to conduct experiments,observe outcomes,and test predictions in real time.Automated assessment and grading:GenAI tools such as ChatGPT and the Intelligent Essay Assessor can reliably review and grade written coursework and provide feedback,thereby ensuring speed,consistency,and objectivity.Various studies have demonstrated that these tools can reduce grading time and deliver accurate and consistent results.GenAI in TransportationGenAI is expected to be one of the primary growth drivers of the global transportation and logistics industry,which is expected to increase at a significant CAGR of 44tween 2023 and 2032,to a value of almost US$19 billion.As the industry grapples with shifting trade flows,margin pressures,rising need for sustainable practices,and increasing demands from shippers and regulators,GenAI offers significant transformative potential.According to a February 2024 global study by IDC,over 50%of transportation companies were already using GenAI with knowledge management,marketing(better shipper/lead conversion,increased dynamic pricing/quoting),and product/service creation,accounting for over 70%of use cases.Another study conducted by Deloitte in July 2024 of over 200 executives found that almost all of them(99%)expect the technology to transform their industry,but over two-thirds(71%)expect this transformation to take more than three years.The transportation use cases witnessing the highest adoption and impact are asset management,route optimization,and warehouse operations.Interestingly,over half of the companies surveyed were found to be running GenAI initiatives within each of these use cases,with around 80%of adopters reporting extremely high or high economic value in each use case.25 25 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Major transportation companies have already started making investments in use cases related to contract consulting,transportation execution,strategy,and customer experience.With the technology still very much in its nascent stages,it promises to disrupt every link in the transportation and logistics value chain over the medium to long term.Key use cases:Route optimization:One long-standing challenge for trucking and freight forwarding companies has been the planning of efficient transportation routes.GenAI models present an opportunity to solve this problem by analyzing data related to tariffs,trade agreements,traffic patterns,public transportation,and other variables to generate optimal routes and minimize costs.One of the main benefits of GenAI in the industry lies in the dynamic optimization of transportation networks in real-time through the analysis of traffic data,pedestrian crossings,and emergency vehicle locations.International shipping companies such as DHL are integrating GenAI models into their processes and analyzing data pertaining to shipment volumes,vessel capacity,and port capacities to determine cost-effective and environmentally friendly delivery methods.Dynamic inventory management:With efficient warehousing of goods key to a successful transportation enterprise,dynamic inventory management assumes critical importance.This is especially true if the volume of goods being handled is large.Therefore,inventory control managers are increasingly using GenAI to analyze data gathered from lead times,demand,stock levels,and other sources,to improve product visibility and prevent stockouts and overstock surpluses.Moreover,GenAI-powered systems can dynamically organize warehouse layouts according to product popularity and order forecasts of certain items,thereby reducing trip time and boosting efficiency.Autonomous vehicles:GenAI can create various realistic virtual driving scenarios to train autonomous cars and advanced driver-assistance systems(ADAS)for unpredictable circumstances.Additionally,the technology can improve autonomous vehicles decision-making abilities by creating simulations of different weather patterns and road conditions.Predictive maintenance and demand forecasting:GenAI can also predict infrastructure and vehicle maintenance requirements before they arise,making it possible for transportation companies to take preventative action and avoid malfunctions and shutdowns.Supply chain managers are increasingly using the technology to analyze historical data related to elements such as seasonality,promotions,customer sentiment,and economic situations.This enables them to create efficient ordering patterns,precisely forecast future trends,and identify hazards.One long-standing challenge for trucking and freight forwarding companies has been the planning of efficient transportation routes.GenAI models present an opportunity to solve this problem by analyzing data related to tariffs,trade agreements,traffic patterns,public transportation,and other variables to generate optimal routes and minimize costs.26 26 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Figure 9:GenAI adoption and impact in transportationHigh adoption,high impactHigh adoption,low impact Low adoption,high impactLow adoption,low impactAdoption too low to measure impactDemand planning70%Percentage reporting“extremely high”or“high”economic value(among companies implementing each use case)Percentage implementing(broad or limited implementing)0P%Note:GenAI in Transportation Survey carried out among 200 executives worldwide,July 2024.Source:Deloitteincomplete-missing info from the source content100Pp00 %Inventory managementFleet managementShipment trackingInternational clearanceWarehouse operationsRoute optimizationAsset managementCustomer serviceFinance and risk managementFrontline workforce productivity27 27 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise AI Infrastructure&ArchitectureAs AI and related technologies continue to evolve,enterprises are making significant investments to develop robust,scalable,and efficient AI infrastructure.According to a 2025 study by S&P Global Market Intelligence,GenAI-related investments exceeded US$56 billion in 2024,almost double from US$29 billion in 2023.An area of interest for investors is the infrastructure layer,which includes semiconductors,graphics processing unit(GPU)cloud,photonic fabrics,high-density compute solutions,edge computing,software tools,and sustainable GenAI infrastructure.Investment in GenAI infrastructure nearly quadrupled in 2024 to almost US$26 billion,up from US$6.86 billion in 2023.The top five GenAI infrastructure trends include:Disaggregated and composable infrastructure:with conventional monolithic architectures becoming expensive and inflexible,enterprises are moving towards disaggregated,software-defined infrastructure,in which compute,storage,and networking resources are dynamically allocated based on workload needs.This includes composable GPU workspaces,particularly in multi-tenant environments,that are fast replacing traditional data centers due to their ability to decouple compute,storage,and networking resources,enabling organizations to reallocate GPU power according to current workloads.For stakeholders,the strategic advantages of investing in composable GPU workspaces include cost efficiency,operational agility,enhanced ROI,and future-proofing IT.Photonic networking for AI Acceleration:the growing size and complexity of GenAI models require ultra-fast,low-latency networking.Cluster sizes are having to quickly scale from just a few AI processors in a server to tens of processors in a single rack and thousands of processors across multiple racks,all while relying on high-bandwidth,low-latency network connectivity to handle huge data transfers.Photonic fabrics are setting new standards for AI clusters,significantly reducing data transfer times and eliminating network congestion.These platforms allow AI compute to be networked seamlessly,from within processor packages to servers across multiple racks.High-density compute solutions:according to recent estimates by Deloitte,continuous improvements in AI and data center processing efficiency could yield an energy consumption level of approximately 1,000 TWh by 2030.These levels of AI workloads demand large-scale hardware infrastructure,making high-density compute solutions critical to achieve maximum output while optimizing power,cooling,and physical space.These solutions are ideal for enterprise GenAI,high-performance computing(HPC),and data center operations.Edge computing:the shift towards real-time AI processing is driving the need for edge computing solutions.GenAI models often require significant computational resources and memory with large model parameters and deep neural networks(DNNs).Edge computing addresses the limitations of traditional cloud-centric architectures by distributing computational resources closer to the data source,reducing latency and bandwidth consumption.AI Industry Trends28 28 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Sustainable infrastructure:GenAI needs massive computational power,rendering it an energy-intensive technology.The production of graphics processing units(GPUs)requires rare earth metals,the mining of which contributes to greenhouse gas(GHG)emissions.Recent estimates suggest that Gen AI could be responsible for creating between 1.2 to 5.0 million metric tons of e-waste by 2030,which is around 1,000 times more e-waste than was produced in 2023.Technology companies are undertaking various initiatives to make GenAI more sustainable.These include energy-efficient chips,smaller models,right-sizing AI/Gen AI workloads,and investments in low-carbon energy sources.A good example is Nvidias new Blackwell chip that has 30 times improved performance for LLM workloads and 25 times lower energy consumption than the preceding iteration.Another example is Googles TensorFlow and Hugging Face,which have incorporated quantization techniques to reduce the size of models,thereby reducing power and resource requirements.Figure 10:GenAI infrastructure funding in 2024Source:S&P Global,as of Jan 10,2025GenAI InfrastructureGenAI Applications20202021202220232024Agentic AI While traditional LLMs are trained on enormous collections of text,images,audio,video,and numbers,and respond to specific human prompts,AI agents(Agentic AI),which build on advanced GenAI models,can act independently,and reason and learn without constant human intervention.Agentic AI technology is gaining traction simply because computers are becoming better at recognizing images and understanding language,mainly due to the evolution of transformer-based technology.Just like humans,these agents work collaboratively using advanced reasoning and planning skills to solve complex,multi-step problems,with LLMs acting as their“brains”for decision-making.What makes them even more attractive is their ability to not only draw from databases and networks but also learn from user behavior and improve over time.Releases such as OpenAIs GPT model family,Anthropics Claude,and Microsofts Copilot are driving the current buzz around Agentic AI.603015450) 29 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Table 6:Agentic AI vs GenAI vs Traditional AIFeaturePrimary FunctionGoal-oriented action&decision-makingAgentic AIContent generation(text,code,images,etc.)Generative AIFocused on automating repetitive tasksTraditional AIAutonomyHigh Operates with minimal human oversightVariable May require user prompts or guidanceLow Relies on specific algorithms and set rulesLearningReinforced Learning Improves through experienceData-driven learning Learns from existing dataRelies on predefined rules and human interventionSource:AISERA2025 and beyondAccording to Maryam Ashoori,Director of Product Management,IBM Watsonx.ai,2025 is expected to be the year when companies begin exploring and deploying agentic AI solutions.An early 2025 U.S.-focused study by IBM and business intelligence company Morning Consult involving 1,000 developers building AI applications for enterprise found that as many as 99%were exploring or developing AI agents.Another study by Deloitte conducted in late 2024 predicted that 25%of the companies that use GenAI will launch agentic AI pilots or proofs of concept in 2025,growing to 50%in 2027.Moreover,some of these applications,in some industries,and for some use cases,could see actual adoption into existing workflows in 2025,especially by the back half of the year.Yet another global study conducted by Capgemini found that 50%of the respondents will implement AI agents in 2025,with the number expected to rise to 82%by 2028.However,Vyoma Gajjar,an AI technical solutions architect,cautions against unbridled optimism,saying that the technologys proliferation requires more than just better algorithms.It needs significant advancements in contextual reasoning and testing for edge cases,and a lack of capabilities in these areas is one of the main hurdles to widespread adoption.Moreover,while the technology is garnering significant attention and investment globally,current Agentic models are prone to making mistakes and getting stuck in loops.In multi-agent systems,“hallucinations”can often spread from one agent to another,which results in a loop of incorrect actions and results.A good example is the AI agent Devin,which was launched by Cognition Software in March 2024 to perform programming jobs unassisted,based on natural language prompts from human programmers.In a recent benchmarking test,Devin was able to resolve nearly 14%of GitHub issues from real-world code repositories,which,even though twice as good as LLM-based chatbots,was still far from being fully autonomous.30 30 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Despite current limitations,the vision of Agentic AI is compelling.The technology is developing at a rapid pace,with some of the latest Agentic AI models employing chain-of-thought functions that,while slower and more deliberative as compared to the more conventional large-scale models,can conduct higher-order reasoning on complex problems.Moreover,multimodal data analysis has the potential to make agentic AI more flexible by increasing the kinds of data that can be analyzed and created.Multimodal AI also shows that agentic AI can be even more powerful when combined with other kinds of AI technologies,such as computer vision(image recognition),transcription,and translation.The global Agentic AI market is estimated to grow from US$7.6 billion in 2025 to US$48 billion in 2030 at a CAGR of 44.5%.Use casesCustomer service:American startups such as Sierra,Ema,and Decagon are developing agentic AI chatbots that can act independently according to their understanding of customer intent and emotions.They operate with multiple specialized agents,each responsible for different aspects of the conversation,such as intent recognition,knowledge retrieval,and emotional understanding.For example,an AI agent could anticipate a delayed delivery,notify the customer proactively,and offer a discount to improve satisfaction.It could also transform customer interaction with conversational support that is empathetic and personalized.Agentic AI chatbots can be of various types:reactive,memory-augmented,tool-using,semi-autonomous,multi-agent networks,and self-improving.Despite limitations,the vision of Agentic AI is compelling.The technology is developing at a rapid pace,with some of the latest Agentic AI models employing chain-of-thought functions that,while slower and more deliberative as compared to the more conventional large-scale models,can conduct higher-order reasoning on complex problems.Figure 11:Global Agentic AI market size in US$billions,2025-2030Source:AgileIntel2025202620272028202920307.611.015.923.033.248.031 31 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Procurement:While current procurement tools focus on data analysis and guided automation,Agentic AI systems such as Zip are already able to function autonomously,guiding employees through complex purchasing decisions by reviewing company policies and requirements.Sales support:Agentic CRMs such as Rox not only store customer data but also help companies get a better understanding of their customers by predicting their needs and proactively engaging with them.U.S.-based 11x has developed two Agentic AI systems,Alice and Mike.While the former functions as a digital sales development representative that autonomously identifies key decision makers and schedules meetings,Mike automates inbound and outbound calls in 28 languages in a personalized,low-latency phone call.Scientific and materials discovery:even though machine learning and non-agentic AI have been used in areas such as drug discovery and new material creation for a long time,Agentic AI is poised to disrupt the field.Agents can not only analyze the properties of specific materials but also propose new materials or combinations based on the characteristics the user is seeking.Moreover,it can also identify optimal suppliers based on priorities such as cost or timing and even order necessary materials.One promising use is ADME(Absorption,Distribution,Metabolism,Excretion)profiling,which predicts drug behavior in the body.A major hurdle is drug candidate failure in later stages due to poor ADME properties or toxicity popping up.Agentic AI can predict these properties early by analyzing molecular structures and historical data,filtering out unfavorable candidates and prioritizing promising ones.Entertainment:Fully autonomous AI agents are already being used in the gaming industry owing to their ability to provide human-like behavior and gameplay for non-player characters(NPCs).For example,researchers created a small virtual town populated with AI by building a sandbox setting similar to The Sims with 25 agents called“Stanford AI Village”.In this village,users can observe and interact with agents as they share news,build relationships,and arrange group activities.Application and Cybersecurity:According to a report by Skybox Security Research Lab,over 30,000 new vulnerabilities were identified in the year leading up to June 2024.As cyber threats grow in number and sophistication,Agentic AI is assuming a critical role in bolstering security postures.This is mainly because the technology outperforms conventional security systems,such as firewalls and antivirus software,to provide a new level of automated defense.It not only analyzes factors like application code,network traffic,user behavior,and system logs to detect anomalies,but also prioritizes these vulnerabilities by risk level and automatically applies patches or recommends fixes.Fully autonomous AI agents are already being used in the gaming industry owing to their ability to provide human-like behavior and gameplay for non-player characters(NPCs).For example,researchers created a small virtual town populated with AI by building a sandbox setting similar to The Sims with 25 agents called“Stanford AI Village”.32 32 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Figure 12:Evolution to multimodal GenAI agentsIntegration of Machine Learning(ML)Learning from data:The integration of ML allowed agents to learn from large datasets,improving their ability to make decisions and perform tasks.This was a significant step forward from rule-based systems,as agents could now adapt to new information and improve over time.Natural Language Processing(NLP)enabled user interactions:Advances in NLP enabled agents to understand and generate human language more effectively,making interactions more natural and intuitive.Introduction of multimodalityCombining text,images,and audio:Multimodal agents emerged,capable of processing andintegrating information from various sources.For instance,an agent could analyze a text description,recognize objects in an image,and understand spoken commands.This multimodality made agents more versatile and capable of handling complex tasks.Enhanced user interactions:Multimodal agents could interact with users in more dynamic ways,such as providing visual aids in response to text queries or understanding context from a combination of spoken and visual inputs.Advanced autonomy and real-time interactionsAdvanced autonomy:Agents can operate independently,rationalize and set their own goals,develop path(s)to attain these goals,and make independent decisions without constant human intervention,leveraging data from multiple sources or synthetic datasets.In a multi-agentic orchestration system,the first set of agents focus on mimicking human behavior(e.g.ChatGPT-4o),that is,thinking fast to come up with solution approach,while the second set of agents focus on slow reasoning(e.g.ChatGPT-1o)to come up with a vetted solution5.Combining thinking fast and slow reasoning,agents can process information and make optimal decisions in real-time crucial for applications like autonomous vehicles,real-time customer service,and various mission-critical business processes.This autonomy makes agentic AI particularly powerful in dynamic and complex real-world environments.User interactions within an ethical and responsible AI-controlled environment:With increased capabilities,there has also been a focus on ensuring that agentic systems operate ethically and responsibly,considering factors such as bias,transparency,and accountability.Source:AgileIntelThe evolution can be broken down into three key phases:2000s2010s2000spresent33 33 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Figure 13:GenAI vs Agentic AI approach to task completionA GenAI Approach to Task CompletionReceiveReceive task or objective from a humanPerceiveProcess input to understand context and gather relevant data(if necessary)GenerateGenerate relevant responses using pre-trained models.Additional Human PromptingHumans interpret the output and then create a new prompt to further iterate on a given task.An Agentic,“Human-like”Approach to Task CompletionReceiveReceive task or objective from a humanPerceive&ReasonProcess input to understand context and gather likely relevant data from various sources.Plan&CoordinateUnderstand,coordinate,and plan tasks to generate useful outputs.Continuous Learning from Environment,Human Feedback&Additional Autonomous Agentic IterationAdapt continuously based on feedback from the environment,refining future responses to achieve target tasks/objectives.ActExecute plans to achieve the task using tools(e.g.,via APIs)Figure 14:Comparative scoring of leading Agentic AI solutionsSource:Cambridge Centre For Alternative FinanceSource:The Futurum GroupSalesforceAgentforce9.5 9.5910TechnicalOperationalFinancialGovernanceMicrosoftCopilot Agents98.5 8.59GoogleCustomerEngagementSuite78.5 8.58IBMwatsonx.aiAgents98.5910Oracle AIAgents87.58.59SAP JouleAgents98.5910DIY/In-HouseDevelopment676.5 6.5ServiceNow AIAgents898934 34 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise AI Governance Risk,Compliance,Responsible AIResponsible AIAccording to a 2024 McKinsey report,global GenAI use doubled in the year leading up to the study,with ChatGPT boasting 200 million weekly active users as of August 2024,double the number from 2023.Another study by Thomson Reuters conducted in 2025 showed that 95%of the respondents believe GenAI to be central to their organizations workflow within the next five years.Pertinently,the pace of GenAI adoption is quicker than that of personal computers and the internet.However,it is the allure of GenAIs potential that has led organizations to dive headfirst into adoption without mitigating risks adequately.This has proven to be particularly challenging given the nascency of the technology as a whole.Moreover,GenAIs greater sophistication as compared to traditional AI poses a huge challenge from a technical standpoint.After all,AI models have evolved from just a few parameters with ML,to tens of thousands with deep learning,and now to millions,billions,and at times trillions with the LLMs.Therefore,companies and organizations are increasingly designing GenAI applications responsibly,addressing potential risks and transparently sharing lessons learned to help establish best practices.According to a 2025 McKinsey report,companies that have been able to capture significant value from the technologys use have consistently paid more attention to address known risks and identify and prevent new ones.According to a 2024 McKinsey report,global GenAI use doubled in the year leading up to the study,with ChatGPT boasting 200 million weekly active users as of August 2024,double the number from 2023.Pertinently,the pace of GenAI adoption is quicker than that of personal computers and the internet.Note:The survey was conducted among business leaders from over 30 countries,N=759Source:Stanford AI Index Report 20251-5M5-10M10-25MFigure 15:Investment in responsible AI by company revenue,202425-50M0 0M1B-10B10B-30B100M-1B30 B100%of respondentsrevenue in USD68%6%1H0%725$0%)!%5 35 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Table 7:Notable RAI policymaking milestonesDateMay 2024OECDStakeholdersGlobalScopeThe OECD updated its AI principles and refined its framework to reflect the latest advancements in AI governance.These principles emphasized building AI systems that take into account inclusive growth,transparency,and explainability,as well as respect for the rule of law,human rights,and democratic values.DescriptionSource:Stanford AI Index Report 2025May 2024Council of EuropeEuropeThe Council of Europe adopted a legally binding AI treaty(The Council of Europe Framework Convention on Artificial Intelligence and Human Rights,Democracy,and the Rule of Law).This treaty was drafted to ensure that the activities within the life cycle of AI systems align with human rights,democracy,and the rule of law.Jun 2024European UnionEuropeThe EU passed the AI Act(EU AI Act),the first comprehensive regulatory framework for AI in a major global economy.The act categorizes AI by risk,regulating them accordingly and ensuring that providersor developersof high-risk systems bear most of the obligations.Jul 2024African UnionAfricaThe African Union launched its Continental AI Strategy(AU AI Strategy),outlining a unified vision for AI development,ethics,and governance across the continent.The strategy emphasizes the ethical,responsible,and equitable development of AI within Africa.Sep 2024United NationsGlobalThe United Nations updated its Governing AI for Humanity report(U.N.AI Advisory Body),outlining efforts to establish global AI governance mechanisms.The report recommends developing a blueprint to address AI-related risks and calls on national and international standards organizations,technology companies,civil society,and policymakers to collaborate on AI standards.Oct 2024G7GlobalThe G7 Digital Competition Communiqu(G7 AI Cooperation)reaffirmed commitments to fair and open AI markets,stressing the need for coordinated regulatory approaches.Previous discussions focused on competition and the regulatory challenges posed by AIs rapid growth.Oct 2024ASEAN and the USAsia and the USFollowing the 12th ASEAN-United States Summit,ASEAN-U.S.leaders issued a statement on promoting safe,secure,and trustworthy AI.They committed to cooperating on the development of international AI governance frameworks and standards to advance these goals.Nov 2024International Network of AI Safety InstitutesGlobalThe first International Network of AI Safety Institutes was established,bringing together nine countries and the EU to formalize global AI safety cooperation.The network unites technical organizations committed to advancing AI safety,helping governments and societies understand the risks of advanced AI systems,and proposing solutions.Feb 2025Arab LeagueArab NationsThe Arab Dialogue Circle on“Artificial Intelligence in the Arab World:Innovative Applications and Ethical Challenges”was launched at the Arab League headquarters,focusing on AI innovations while placing a strong emphasis on ethical considerations.Responsible AI(RAI)is a comprehensive and holistic framework that guides companies and other organizations to implement AI in a way that enables them to benefit from AI systems while mitigating risk and remaining consistent with corporate values.For GenAI to be integrated across industries at scale,companies must implement the principles of RAI across the full application life cycle by governing their data,protecting company intellectual property(IP),preserving user privacy,and complying with laws and regulations.One way of doing it is by automating and scaling parts of AI governance,security,and risk management programs to detect and monitor configured guardrails and controls more efficiently.Another way is to adopt a risk-tiered approach that applies different monitoring standards to AI systems based on risk and impact on customers,partners,and employees.36 36 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise WestRocks GenAI integration has yielded higher productivity and lower costs.Paul McClung,VP of internal audit at WestRock,a global sustainable,fiber-based packaging solutions company,first heard about GenAI in 2022 but dismissed its use to augment the companys audit function.However,the companys IT department developed a secure GenAI platform in late 2023 for all internal departments to experiment with.One of the first applications of the technology was on the front end of the audit process to draft objectives.When this proved to be successful,Paul decided to automate the entire audit process by ingesting data and running it through a seamless model with a click of a button.However,the team found that linking several tasks instead of executing them individually would prove to be more effective.Another effective strategy was to integrate a high level of standardization within internal processes,starting with standard prompts to write audit objectives and execution methods.This enabled WestRock to automate the process of creating sample risk and control matrices,draft audit programs,and even suggest technology tools and scripts for the company to consider.Some of the early value captured through the use of GenAI has unsurprisingly been higher productivity and lower costs.However,Paul cautions that these benefits are still very much in their nascent stages,and to realize optimum value,the company will have to fully reengineer its processes,timelines,milestones,and resource deployment models.It will also have to move away from its previous strategy of getting its programmers to develop scripts based on requirements,to a more iterative process that involves developing scripts in real time and adjusting as needed.This requires a team approach where multiple people challenge the results of the GenAI models,but in a condensed time frame based on the technologys speed.According to Paul,WestRocks future with GenAI technology involves integration with Agentic AI to add a learning mechanism to the platform that builds on historical lessons to improve and expand its scope of operations in the future.Another immediate goal is to leverage the companys learnings with data analytics to improve the implementation of GenAI.This includes leveraging the platform with continuous monitoring and full population assessments rather than just sampling.GenAI in Enterprise:Case Studies WestRocks future with GenAI technology involves integration with Agentic AI to add a learning mechanism to the platform that builds on historical lessons to improve and expand its scope of operations in the future.Another immediate goal is to leverage the companys learnings with data analytics to improve the implementation of GenAI.37 37 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Over the medium term,the company is expected to develop a dynamic system in which risk assessment questions are generated based on changes in industry and external environment data gathered and analyzed in real time.This will likely result in follow-up reporting,action tracking,and trending being automated through interactive chatbots.McDonalds much-touted conversational AI solution was withdrawn from the U.S.in June 2024McDonalds acquisition of voice-based conversational AI technology company Apprente in 2019 marked the beginning of the companys exploration with GenAI.Apprente specialized in developing sophisticated speech recognition and natural language processing(NLP)systems designed to handle complex,multi-lingual,and context-sensitive interactions.These solutions were expected to automate McDonalds drive-thru systems and streamline the order-taking process.In October 2021,the company forged a strategic partnership with IBM to leverage its AI and cloud computing expertise to expand the deployment of AI-powered drive-through systems across more locations.However,despite the integration of sophisticated technology,the GenAI-enabled system frequently misunderstood customer orders with background noise,varied accents,and complex orders,leading to significant misinterpretations.In fact,many videos of the AIs failures were recorded and widely shared on social media,causing much negative publicity for McDonalds.In June 2024,the fast-food chain withdrew the automated systems from over 100 locations around the U.S.The key reasons for failure are mentioned below:Real-world testing:One of the main reasons for this failure was the lack of real-world testing to ensure the systems can handle the variability of actual customer interactions.This includes simulating different accents,background noises,and complex order scenarios.Moreover,the system wasnt trained on exhaustive and diverse datasets that were updated regularly to keep it adaptable to new linguistic patterns and customer behaviors.User-centric design and feedback loops:The company failed to incorporate user feedback into the development cycle to continually refine and improve the system.This is especially important for a company like McDonalds,in which understanding user needs and expectations is crucial for designing AI systems.AI systems should be continuously updated and refined based on real-world performance data and user feedback.Establishing feedback loops allows for ongoing improvement and adaptation to changing conditions and user behaviors.This iterative process helps maintain the systems relevance and effectiveness over time.One of the main reasons for this failure was the lack of real-world testing to ensure the systems can handle the variability of actual customer interactions.This includes simulating different accents,background noises,and complex order scenarios.38 38 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise The latest LLMs such as GPT-4(1.8T parameters),Claude 3(2T parameters),and Metas LLaMA 3(405B parameters),are now being trained on billions,or even trillions of parameters,resulting in significant advancements in natural language understanding,code generation,and reasoning.In fact,some of these models are now operating at or near human-level accuracy on functions such as reading,image recognition,speech recognition,and language understanding.Some of the top current LLMs include:Claude:created by Anthropic,Claude focuses on constitutional AI and has three primary branches-Opus,Haiku,and Sonnet.Its latest iteration is the Claude 3.5 Sonnet that can decipher nuance,humor,and complex instructions better than previous versions.The LLM also has broad programming capabilities that make it ideal for application development.In October 2024,Claude added a computer-use AI tool that allows it to use a computer like a human does.DeepSeek-R1 is an open-source reasoning LLM that uses reinforcement learning to deliver mathematical problem-solving and logical inference capabilities.DeepSeek-R1 can perform critical problem-solving through self-verification,chain-of-thought reasoning,and reflection.Ernie:released by Chinese technology company Baidu in August 2023,Ernie is said to have 10 trillion parameters and has garnered 45 million users globally.Gemini:a product of the Google family of LLMs,Gemini models are multimodal and available as a web chatbot,the Google Vertex AI service,and via API.They are available in three variants Ultra,Pro,and Nano.Ultra is the largest and most capable,Pro is the mid-tier model,and Nano is the smallest model,designed for efficiency with on-device tasks.The latest version of Gemini,the Gemini 1.5 Pro,was released in May 2024.Llama:developed by Meta,Llama was first released in 2023 and then subsequently in July 2024 as both a 405 billion and 70 billion parameter model.The most recent version is Llama 3.2,which was released in September 2024,initially with smaller parameter counts of 11 billion and 90 billion.Llama uses a transformer architecture and was trained on many public data sources,including webpages from CommonCrawl,GitHub,Wikipedia,and Project Gutenberg.ChatGPT continues to be the market leader,but its growth has slowed as Google and Microsoftintroduce enhancements to their AI assistants.Among startups,general-purpose AI chatbots areexperiencing gradual but consistent user acquisition,while business-focused Claude AI is currentlyleading in terms of growth.GenAI Technology 39 39 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Table 8:Significant model and dataset releasesFeatureDeveloperMicrosoftCopilot(Microsoft)OpenAIChatGPT(OpenAI)Google DeepMindGemini(Google)Source:Swiss German University,Web SearchMeta AILlama(Meta)Latest ModelMicrosoft 365 Copilot(2025)GPT-4.5(2025)Gemini 2.5(2025)Llama 4(2025)Primary FocusIntegration of AI in Microsoft appsGeneral AI,conversation,codingMultimodal AI,Google ecosystem-Training DataBuilt on OpenAIs GPT-4,ProprietaryBroad,multimodal,diverseWeb-scale,multimodal-Key FeaturesDeep integration with Microsoft 365 and GitHub Web browsing,DALL-E,document analysis,and voice interactionsMultimodal(text,images,audio,video),Google services integrationOpen-source LLM optimized for research and on-device deploymentCode GenerationExcellent in Python,JavaScript,C ,JavaExcellent(Python,JS,SQL)Strong(Python,JS,SQL)Average Multimodal SupportSupports text and image generationStrong(images,text)Very strong(text,images,audio,video)Text-only;no native multimodal support for images,audio,or video.Memory FeatureYesYes(for Plus users)YesYesAPI AvailabilityNoYesYesYesFree VersionYes(Microsoft Edge)Yes(GPT-3.5)Yes(Gemini 1.0)Yes(Llama 2 and Llama 3.2)StrengthsCode-specific assistanceVersatile,reliable general AI,Strong conversational abilities,Integrated with pluginsBest multimodal AI,Google ecosystem integration,Strong reasoningPrivacy&mobile deploymentWeaknessesOver-reliance risksNo real-time browsing in the free version,can generate hallucinations,and Advanced features are behind a paywall.Requires Google integration,some accuracy issuesLimited complexity handlingBest ForSoftware developmentGeneral-purpose AI,chatbots,writing,and researchMultimodal tasks,search,and productivityOffline,low-resource environmentsCost StructureIncluded with Microsoft 365 subscriptionsSubscription-based(Plus/Team)Free with a Google accountSubscription-based40 40 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Table 9:Leading GenAI models and specificationsModelCommand-RCohereCreator128kContext Window15Artificial Analysis Intelligence Index6.97End-to-End Response TimeJamba 1.6 MiniAI21 labs256k182.89DBRXDatabricks33k20NACodestral(May 24)Mistral AI33k205.01LFM 40BLiquid32k223.23Qwen3 0.6B Alibaba32k23NAYi-LargeAlibaba32k287.78Nova MicroAws130k281.79Tulu3 405BAi2128k40NAPhi-4MS Azure16k4012.98Phi-4MiniMax4m4016.51Sonar ProPerplexity200k437.98Reka Flash 3Reka128k4745.06Claude 3.7 SonnetAnthropic200k487.44GPT-4o OpenAI128k503.83Llama 4 MaverickMeta1m514.21Grok 3X.AI1m5110.36DeepSeek V3Deepseek128k5322.47Gemini 2.5 ProGoogle1m6839.73OpenChat 3.5Openchat8kNA10.87ArcticSnowflake4kNANASolar MiniUpstage4kNA38.52Note:Context windows-Maximum number of combined input&output tokens.,Artificial analysis Intelligence Index-a comprehensive benchmark used to evaluate and compare the intelligence of language modelsSource:Artificial Analysis41 41 Visit SiteVisit SiteRegisterState of Generative AI in the enterprise Table 10:Illustrative capabilities of GenAI platforms from select frontier labsAnthropicClaude Not multimodal(text only)Limited contextual understanding(difficulty withcomplex conversations)No tool usage2022-23Jan 2025Claude 3.5 Multimodal(text,audio,and images)Enhanced contextual understanding and coherence during long interactions Experimental computer usage capability for some usersGoogleGeminiGoogle Bard Not multimodal(text only)Fair reasoning Limited contextual understanding(difficulty withcomplex conversations)Limited real-time data integration Low personalization(limited adaptability)Gemini 2.0 Flash Multimodal(text,audio,and images)Advanced reasoning(capable of multistep problem-solving and nuanced analysis)Enhanced contextual understanding(maintains coherence in long dialogues)Real-time data integration(from Google Search)Advanced personalization(user context)MetaLlama 1 Not multimodal(text only)Fair reasoning Limited contextual understanding(difficulty withcomplex conversations)No API accessLlama 3.3 Text-based(earlier versions were multimodal,LLaMa 3.2)Advanced reasoning(capable of multistep problem-solving and nuanced analysis)Enhanced contextual understanding(maintains coherence in long dialogues)API access(tools for model and agent development)MetaMicrosoft Phi-1 Not multimodal(text only)Fair reasoning(i.e.,limited to coding tasks)Focused training(smaller,coding-focused data set)Phi-4 Multimodal(text,audio,and images)Advanced reasoning(capable of multistep problem-solving and nuanced analysis)Comprehensive training(diverse data)OpenAIGPT-3.5 Not multimodal(text only)Fair reasoning ability(e.g.,scored high on SAT,butbottom 10%on bar examination)Limited contextual understanding(difficulty withcoherence in complex conversations)Standard API access(for text generation)OpenAI 0 Multimodal(text,audio,and images)Advanced reasoning(e.g.,top 10%on bar examination)Enhanced contextual understanding and coherenceduring long interac
2025-10-29
60页
5星级
IBM Institute for Business Value The State of Salesforce 2025-20263 pillars for cost-effective agentic AI at scaleHow IBM can helpPartnering with Salesforce for more than 25 years,IBM has developed a deep understanding of the unique needs of large enterprises.By combining our expertise in AI,hybrid cloud,and enterprise transformation with the power of Salesforce,we help unlock new levels of efficiency,innovation and growth on your Salesforce journey.Our certified Salesforce consultants solutions achieve customer success by harmonizing your existing infrastructure,minimizing disruption and maximizing ROI on your CRM investment.For more information,visit https:/ 5Pillar 1:Optimize ROI while managing unpredictability 8Pillar 2:Make governance your competitive accelerator 14Pillar 3:Give agents a 360-degree view beyond Salesforce 20Action guide 26Contents12Despite rising AI investments,74%of Salesforce customers still struggle to move the needle on customer experience and engagement.The ROI on Salesforce AI investments depends on where,and how,the technology is used.Agents and assistants show early traction in IT,sales,and service,proving their value and justifying continued investment as cost management challenges rise.Only 21%of Salesforce customers are confident they have the right governance for agentic AI.As digital labor increasingly makes autonomous decisions,especially in customer-facing areas,most organizations fall short on the accountability necessary for human-AI collaboration.Only 26%say most of their customer data resides in Salesforce,which means nearly three out of four Salesforce customers are needlessly missing a treasure trove of value.Those connecting Salesforce to third-party platform and mainframe data see better business outcomes,including deeper customer insights and cost savings.3Pillar#1:Optimize ROI while managing unpredictability.Pillar#2:Make governance your competitive accelerator.Pillar#3:Give AI agents a 360-degree view beyond Salesforce.Key takeaways4|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide5 5Last years State of Salesforce highlighted three priorities:unlocking the full potential of data,tailoring AI use cases for optimum business impact,and building competencies through ecosystem partners.1 The state of the market was clear companies werent held back by a lack of data,but by the ability to leverage existing data effectively.Fast forward to today,and the challenge has intensified.The AI revolution continues to promise transformation,but for many Salesforce customers,its delivering more invoices than insights.Still caught among complex tech stacks and fragmented data architectures,they face mounting pressure to deliver ROI on AI investments.New research from the IBM Institute for Business Value(IBM IBV)highlights the growing gap between ambition and reality.For 1,200 Salesforce customers surveyed,only 33%of AI initiatives are meeting ROI targets.Even more concerning:72%have failed to scale across business units,and 20%have stalled,failed outright,or been abandoned.When customers demand instant relevance and precision from brands,2 organizations simply cant afford to slow down.They must figure out how to turn AI from a cost center into a financial growth engine.This underperformance comes at a pivotal momentjust as AI makes its evolutionary leap from advisor to actor.While traditional AI tools suggest and automate narrow tasks,requiring constant human supervision,a new breed of autonomous agents is emerging.These agentic systems independently perceive,reason,and execute complex actions toward defined goals with minimal human oversight.Salesforces introduction of low-code agent builders has democratized this advanced capability,and executives are responding with renewed optimism.When asked about specific builders such as Agentforce,most respondents in our study are betting big,predicting it will boost operational efficiency(68%),fuel product innovation(65%),and drive business outcomes(61%).Agentic AI is bundled in with CRM platforms but its on leaders to deliver the ROI|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide6But heres a catch:even though agent builders are bundled into Salesforce and other platforms,business value isnt guaranteed.Without a shared source of truth across clouds,platforms,functions,and workflows,agents will be limited to isolated tasks,unable to deliver value across the enterprise.Our research shows that unlocking outcomes with agentic AI takes more than turning on features or building agents.It demands a reimagining of how AI,platforms,and people come together to serve the customerin every moment,on every channel.Our analysis revealed a group of organizations that are most likely to successfully scale AI agents across the enterprise.They have optimized the ROI of AI amid cost unpredictability,prioritized rigorous governance,and harnessed a gold mine of data beyond Salesforce.These leaders report 60%greater efficiency,57%more effective customer insights,and more than two times greater pipeline expansion.Their AI investments not only deliver expected ROI more frequently than their peers but also enable scalability across the business.In this report,we explain how enterprises can follow these pioneers and bridge the gap between AI promise and performance.In part one,we explore financial optimization,showing how organizations can accelerate AI ROI and manage cost unpredictability through strategic deployment.Part two addresses the critical need to evolve governance models to match AIs increasing autonomyembedding guardrails that build confidence and enable faster scaling.In part three we focus on the interoperability of Salesforce with other systems,demonstrating how enterprise-wide data sources can deliver greater value.We end with an action guide,offering specific steps organizations can take now to turn agentic AI potential into tangible results.Salesforce customers say only 33%of AI initiatives are meeting ROI targets.|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide7 7Negative impact No impact Positive impact Operational efficiencyProduct and service innovationCustomer experience 8%h#e%8p%Biggest barriers to agentic AI adoptionData availability and qualityUnclear business valueFinancial viability419Y%of Salesforce customers say the benefits of agentic AI will outweigh the effort and investment required to scale it successfully.H He er re e is a snapshot of where they are.Areas of agentic AI adoption32%IT,security,or data management 21%Human resources31%Customerservice 20%Operations and supply chain19%Business development24%Sales20%Marketing 53%Expected impact of agentic AI on key business outcomes Agentic AI outlookPerspective|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide8Give agents an outcomeor take them off the payroll.Pillar 1:Optimize ROI while managing unpredictability|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide9The AI gold rush has left executives with a distressing invoice.AI spend is growing rapidly,yet key functions face sprawling ecosystems of clouds and tools that dont fully communicate.What should be powering growth is instead a financial quagmire.The shift to consumption-based AI pricing is only turning up the pressure.In our research,62%of Salesforce customers express concern about unpredictable AI-related costs,and 64%report that unclear total cost of ownership makes it difficult to determine whether AI agents will save money or quietly inflate budgets.The problem extends beyond simple price tags to include implementation timelines,maintenance costs,continuous retraining,platform updates,and the often-overlooked expenses of employee upskilling as well as adoption and performance tracking.These all add up to the inability to measure and manage AIs true financial impact over time.64%of Salesforce customers say unclear total cost of ownership makes it difficult to determine whether AI agents will save money.|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide10Forward-thinking organizations fight financial and operational uncertainty by treating AI agent investments with the same disciplined oversight they apply to human workforce decisions.Rather than relying solely on projected savings or revenue,leaders can use risk-informed ROI frameworks that consider both holistic financial returns and broader organizational impacts to help create clarity amid the complexity.Executives also need accountability for where and how AI agents are deployed(see Perspective,“How to measure AI agent performance”).That means tying each agent to specific business outcomes,setting baselines up front,and measuring results against both financial and operational expectations.And,importantly,it requires a willingness to sunset what doesnt deliver.Organizations report growing investment in areas where AI can directly impact both the top and bottom line(see Figure 1).Customer self-service leads the way(70%)followed closely by e-commerce(65%).Back-office operations are seeing a parallel surge:64%point to increasing investments in marketing automation,61%in IT platform and service management,and 57%in security and compliance monitoring.Top organizations also match their AI approach to the use casenot the other way around.While agentic AI offers advanced autonomy,it isnt automatically the most effective choice for every scenario.Different types of AI should be used as appropriate.Today,Salesforce customers report generative AI is delivering the strongest ROI in marketing,sales,and IT,while AI agents and assistants are gaining early traction in IT,sales,and customer service.Aligned with business needs,these targeted applications are proving their valueand earning sustained investment in an uncertain economic landscape.How to combat AI cost uncertaintyWhere AI is proving its worth|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide11Back-office AI investmentsMarketing automation64%IT(platform and service management)61%Security and compliance monitoring57%Revenue operations 55%Field service53%Workforce planning49%Supply chain36%Front-office AI investmentsCustomer self-service70e%E-commerceSales enablement63%Contact center support&service56%Distribution channels51%Telemarketing50%In-person direct sales44%AI is moving beyond front-office experiments to reshape the entire enterprise.Percent saying investments have somewhat or significantly increased because of AIFigure 1Q.How has AI changed your organizations financial investments in the following areas/revenue channels over the last year?11|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide12Holding AI accountable is easier said than done.Unlike human employees,AI agents dont submit reports,sit in performance reviews,or visibly“clock in.”That makes measuring their impact trickyand its why many organizations struggle to define success beyond vague aspirations such as“improved efficiency”or“better experiences.”A practical starting point:give every AI use case a“job description”tied to risk-adjusted ROI.What is this agent supposed to accomplish,and what is the performance baseline?Crucially,what are the expected benefits after accounting for uncertainty,operational risk,and adoption rates?What will success look like at 30,90,and 180 days?In sales,for example,an AI assistant might improve lead conversion or reduce days-to-close,but the evaluation should weigh potential revenue gains against the risk of incorrect recommendations or failed automations relative to the baseline.In customer service,an agent could resolve cases faster,but metrics must adjust for escalations or human intervention.These arent just metrics;theyre accountability anchors that allow executives to measure not just raw performance,but risk-adjusted impact.Without them,AI stays in the realm of“cool tech.”With them,it becomes a measurable contributor to business performance.The right metrics will vary by function and use case.But the principle holds:if youre going to put AI on the payroll,make sure it has a joband a way to prove its doing that job well.How to measure AI agent performancePerspective12|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide1313|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide14Workflow decisions are being made with or without you.Pillar 2:Make governance your competitive accelerator14Workflow decisions are being made with or without you.|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide15AI systems are increasingly making autonomous decisions affecting customers and operations.With operational integrity and brand reputations on the line,how decisions are made,security,and risk management must be more than a compliance checkbox.But the adoption of governance frameworks isnt keeping pace with technology advancements.Nearly three in four Salesforce customers say the increasing use of digital labor elevates the need for risk management,which they also report as the second most critical success factor for agentic AI initiatives.Yet only 21%strongly agree that they have the governance necessary.A mere 9%consider risk a top consideration when evaluating AI tools and initiatives.The governance challenge extends to talent management.Two-thirds of leaders expect agentic AI to change how they structure teams and departments,and most executives(73%)anticipate AI will free employees for higher-value tasks.But they are rushing into a new era of human-AI collaboration without the necessary guardrails.In fact,only 30%of executives consider clear guidelines around human-AI collaboration a critical success factor for agentic AI initiatives.And more than half(56%)have not issued comprehensive guidance to employees on the use of AI.This could explain why executive confidence in using AI to transform customer and employee experiences remains frozen at last years level,16%,even as cautious optimism grows(see Figure 2).3 The governance gap presents a strategic opportunity:organizations that establish trusted governance gain the confidence to accelerate AI adoption and scale responsiblyultimately allowing them to innovate faster and seize a competitive advantage.In our research,organizations further along in scaling agentic AI also had strong governance practices in place,and consequently had greater confidence in transforming experiences with AI.How governance boosts confidence in AI|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide16Executives shift from concern to cautious optimism about their AI capabilities,but full confidence remains elusive.Figure 2Q.What best describes your organizations view about using AI workflows to transform customer and employee experiences?Confident16utiously optimistic41%Interested,but concerned about implications20%Interested,but dont know where to start 17%It will not bring the value we need 3%We are not ready 3%7 25202416|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide17If governance is so critical,why has implementation lagged?Because true governance transcends policiesit requires anticipating how AI systems behave under real-world pressures,in varied contexts,and over time.Without this foresight,blind spots multiply:corrupted or misleading data stored in an agents memory slowly distorts decisions,poorly scoped integrations trigger unintended actions,and unchecked privileges open the door to security risks.4Industry and geographical realities compound these challenges.Healthcare and financial services organizations navigate strict compliance rules.European entities face data sovereignty and cross-border requirements.Meanwhile,companies integrating AI with mainframe systems face elevated security and privacy considerations for their most sensitive data,and the shift toward more open AI modelsopen source,open weights,open dataintensifies the demand for transparency and auditability.In this environment,governance must evolve from a compliance function to the operating system for enterprise AI.Traditional approaches built on static policies and periodic checkpoints cant match agentic AIs speed and adaptability.Instead,governance must become dynamic and embeddedsetting guardrails that help prevent costly errors while enabling growth,rapid decisions,and differentiated customer experiences.Compliance,security,and risk thresholds must be built directly into AI agents so they can operate safely in regulated workflows without slowing the businesssupported by continuous feedback loops to detect drift early and by clear protocols for human intervention.Organizations that embrace this proactive approach can make risk mitigation and governance the foundation for responsible,high-impact AI adoption and a strategic differentiator.Closing the gap in AI governanceGovernance must evolve from a compliance function to the operating system for enterprise AI.|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide18Case studyNDS Cognitive Labs and Clip:Expanding financial inclusion in Mexico through AI5Resolves over 66%of inquiries autonomouslyDelivers 80ster average interaction times compared to traditional call centersHas driven a 28%annual increase in chatbot usage.Clips omnichannel service can handle 6,000 simultaneous interactions while freeing human agents to focus on complex issues.The results are impressive.The virtual assistant:Mexico faces significant financial inclusion challenges,with 38%of adults remaining unbanked despite owning mobile phonespresenting an opportunity for digital financial services.Clip,a rapidly growing Mexican fintech unicorn,is helping advance financial inclusion by enabling hundreds of thousands of businesses,many of whom are new to financial services,to accept digital payments.With thousands of new users monthly and without traditional banking infrastructure,Clip needed an efficient way to support customers unfamiliar with financial services concepts such as fees,account setup,and credit lines.NDS Cognitive Labs customized and deployed an AI virtual assistant platform tailored to Clips specific needs.NDS Cognitive Labs used natural language processing,integrated with NeuralSeek and running on the cloud,to develop an advanced conversational AI solution that leverages retrieval-augmented generation(RAG).This approach allows the chatbot to identify user needs and retrieve real-time information from external databases,providing contextually rich responses to a wide range of inquiries.The solution seamlessly integrates with Salesforce Cloud,enabling the AI assistant to generate leads,create support cases,validate identities,and access CRM data.Through natural language processing,it recognizes product references and guides customers with suggestions when needed.18|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide1919|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide20Agents arent the starting point.Your data is.Pillar 3:Give agents a 360-degree view beyond Salesforce20|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide21Because AI agent creation is low-code,teams can be tempted to begin building right away.But starting here is riskydata must come first.On the competitive battleground for customer loyalty,data powers what matters most,and the stakes are high.In our Salesforce customer survey,customer experience(66%),product and service innovation(60%),and AI adoption(57%)top the priority list over the next two years.And 69%of executives now say AI must be interoperable across platforms to have significant impact on their business.Awareness doesnt always lead to outcomes.Despite rising AI investments,74%of Salesforce customers still struggle to move the needle on customer experience and engagement.Why?Because delivering on customers new standards for precision,prediction,and protection depends on the right technology foundationwhich remains a challenge for many.When asked about their biggest roadblocks,Salesforce customers cite modernization of legacy systems(64%)as their number one concern.For agentic AI specifically,poor data availability and quality is the leading adoption barrier(53%).Why fragmented data persists as a barrier to AI69%of executives say AI must be interoperable across platforms to have significant impact on their business.|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide22The fundamental issue is fragmented data.Low-code platforms accelerate the building of AI agents,but for most organizations,these agents are still flying blind,cut off from reservoirs of enterprise data.Only 26%of executives report that their customer data primarily lives within Salesforce.For the remaining 74%,critical data that contains competitive advantagefrom order histories in ERP systems to claims data in insurance mainframes to transactional and billing data in financial systemsis trapped in silos,effectively limiting the view of AI agents into the full customer context.Organizations that ignore these data gaps are 45%less likely to have successfully implemented agentic AI in their enterprise.For agentic AI specifically,poor data availability and quality is the leading adoption barrier(53%).How integration technology dissolves data silosFortunately,tapping into the data held in existing systems doesnt require rebuilding from scratch.Modern integration technologies such as zero-copy data integration and data fabrics make it possible to unleash valuable data from complex back-office systems and mainframes,without the cost,disruption,or risk of moving it.Instead of physically migrating data,zero-copy creates virtual access points so applications and analytics tools(including AI)can work with information in real time while leaving data securely where it lives.Data fabric technologies then connect these sources across silos.And emerging AI interoperability standards such as Model Context Protocol(MCP)and AI-to-AI(A2A)which now incorporates the Agent Communication Protocol(ACP)establishes the framework for seamless system integration and communication.In banking,integration technologies can enable pulling fraud-detection signals from credit systems without migration.In healthcare,it could involve combining electronic health records with scheduling and billing in real time,without touching the underlying systems.|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide23The proof is in the performance(see Figure 3).Organizations that connect Salesforce with external platforms such as Adobe,SAP,or Oracle are nearly three times more likely to report a highly successful 360-degree customer view,2.4 times more likely to produce cost savings,and twice as likely to cite deeper customer insights.These benefits translate into broad business outperformance.Connected enterprises report significantly stronger revenue growth,profitability,innovation,and customer engagement than their peers.For a$19 billion company,connected platforms could mean roughly$140 million in additional revenue.Plus,it can mean a 2%savings in reduced implementation and development costs.Even mainframe integrationoften seen as complex and high riskcan deliver substantial returns.Salesforce customers that have connected mainframe data were nearly 30%more likely to report significant cost savings and more accurate AI predictions when compared to those without mainframe connectivity to Salesforce.In the end,an AI agent is only as effective as the data it can access.The future of customer experience will be won by those who build connected,intelligent ecosystems where technology,data,and human insight converge.23|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide24Case studyWater Corporation:Transforming concession processing with integrated data7Western Australias Water Corporation delivers essential water services to over 1.3 million customer accounts,including processing more than 40,000 annual concession applications for pension recipients and other eligible customers.Their decade-old application processing system relied on manual workflows that couldnt scale to accommodate growing workloads and struggled to store historical records securely.Partnering with IBM,Water Corporation implemented an eConcessions platform powered by Salesforce Service Cloudextending its use of Salesforce as their strategic customer support platform.The eConcessions platform offers streamlined,automated process workflows that seamlessly transition concession applications throughout the initial stages of validation and authorization.The solutions integration capabilities allow data exchanges in near real time,reducing human errors that could delay a given application.The results were transformative:The corporations existing websiteBilling and customer communication systemsLocal government authority systems for external processing.The extensive integration capabilities of the Salesforce platform enable the solution to pair with:Automated concession verification in most casesAccelerated validation processes Improved employee efficiency and satisfaction through consolidated data access in a single consoleEnhanced scalability to accommodate growing application volumes and historical records.24|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide252.4xmore accurate AI predictions 2.7xmore likely to report a highly successful 360-degree view of the customer2.4xmore likely to report a significantly positive impact on customer insights2.6xmore likely to report a significantly positive impact on prediction accuracy2xmore likely to successfully drive cost savings with SalesforceConnecting external platform data to SalesforceConnecting mainframe data to SalesforceBased on IBM IBV analysis.Organizations breaking down data silos realize better business outcomes.Figure 325|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide26Make ROI a prerequisite not a postmortem.0102Elevate risk management from checkbox to growth enabler.Agentic AI is ultimately a new way of doing business.Salesforce customers have access to a powerful platform tool to harness the power of AI agents.But outcomes are earned,not given.Unlocking the full ROI of agentic AI demands a strategic reimagining of how data,systems,and people work together.For enterprises willing to invest in the effort,the payoff is significant:seamless customer experiences,operational efficiency,and innovation at scale.For everyone else,even the most advanced AI agents will perpetually fall just short of true impactfull of promise,but unable to deliver the transformative outcomes executives expect.Here are three actions that can help you convert potential into reality.Review your AI portfolio to identify initiatives that deliver the strongest returns for customers,employees,and the business,and start with sales,customer service,or IT security for agentic use cases.Establish a standardized framework baseline performance before deployment.Measure ROI after deployment,factoring in risk-adjusted business benefits as well as data usage and effectiveness.Launch initiatives quickly to test,reassess,and mitigate risk,creating a cycle of rapid learning and optimization.By orchestrating agentic AI for both user value and business outcomes,organizations can optimize portfolio returns and accelerate results.Embed governance throughout the AI lifecycle.Integrate governance checkpoints across the entire AI journeyfrom initial design and training to deployment,monitoring,and scaling.Create cross-functional governance teams that include both technical and business stakeholders to provide balanced decision-making about AI agent risks and benefits.Implement continuous monitoring tools that provide real-time visibility into AI agent activities and decisions,and assign accountability,enabling proactive risk management.Use strong governance as a competitive differentiator to confidently scale AI faster,with trust intact.Turning potential into payoffAction guide26|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide27Mine your systems for data to fuel agents with a comprehensive customer view.Go beyond basic integrations by designing a composable strategy that simplifies architecture,consolidates platforms,and tightly aligns Salesforce with core enterprise systems and data infrastructure.Leverage zero-copy integration and data fabrics to reduce latency,avoid costly duplication,and accelerate agentic AI readiness.Thoughtful integration unlocks richer insights,improves accuracy,speeds decision-making,and drives measurable cost savingswithout expensive rip-and-replace projects.03Leave no data behind.27|Introduction|Pillar 1|Pillar 2|Pillar 3|Action guide28Bethany Perkins Former Go-to-Market Strategy LeaderCody McKinney IBM CHQ MarketingArvind Raj Strategy and Transformation Consultant IBM Institute for Business ValueSrinivas RudrabhatlaEMEA Partner,Salesforce Service Area Leader IBM Consulting Kohli Global Research Lead,Customer Experience Transformation IBM Institute for Business Value 25 years of industry and consulting experience in sales,service,marketing,and commerce,Srini has worked in leadership roles with multiple organizations in the EMEA region.He is passionate about helping organizations in strategic transformation and change.Srini is recognized for his experience in customer relationship management,consumer experience,and AI,bringing a human-centric mindset and deep expertise in industry processes,delivery,and technology platforms.With 10 years of experience in product management and strategy,Nisha leads customer experience transformations that connect strategy,design,and technology.She partners with clients to reimagine how digital products serve customers,employees,and the businessdelivering streamlined experiences that drive growth.Sara Aboulhosn Associate Creative Director,IBM Institute for Business ValueBenjamin Hurte Design Director,IBM ConsultingJoanna Wilkins Editorial Lead,IBM Institute for Business ValueWith over 30 years of experience,Tim drives agile,AI-powered transformations that deliver measurable client outcomes.He combines strategic vision with hands-on expertise to shape,sell,and deliver complex projects.Tim ElcottPartner,Salesforce Go-to-Market Leader,UKI Salesforce Practice IBM Consulting RoweSenior Partner,Global Salesforce,Service Area Leader IBM Consulting is an accomplished executive with over 30 years of experience in scaling customer relationships,driving growth,and leading global technology organizations,including key roles at Salesforce,Yext,and Accenture before joining IBM.She leverages her deep expertise in Salesforce to help clients navigate digital transformation and complex industry challenges.29IBM Institute for Business ValueFor two decades,the IBM Institute for Business Value has served as the thought leadership think tank for IBM.What inspires us is producing research-backed,technology-informed strategic insights that help leaders make smarter business decisions.From our unique position at the intersection of business,technology,and society,we survey,interview,and engage with thousands of executives,consumers,and experts each year,synthesizing their perspectives into credible,inspiring,and actionable insights.To stay connected and informed,sign up to receive IBM IBVs email newsletter at can also follow us on LinkedIn at https:/ibm.co/ibv-linkedinThe right partner for a changing worldAt IBM,we collaborate with our clients,bringing together business insight,advanced research,and technology to give them a distinct advantage in todays rapidly changing environment.Subscribe to our IdeaWatch newsletterJust the insights.At your fingertips.Delivered monthly.Brought to you by the IBM Institute for Business Value,ranked#1 in thought leadership quality by Source Global Research for the second consecutive year.Research-based thought leadership insights,data,and analysis to help you make smarter business decisions and more informed technology investments.Subscribe now:ibm.co/ideawatch30Study approach and methodologyThe IBM Institute for Business Value(IBM IBV),in cooperation with Oxford Economics,conducted a global survey in the second quarter of 2025.The survey targeted Salesforce customerswith a total sample size of 1,222 respondents.The study spanned 23 geographies,maintaining proportional geographic representation.It included leaders from 22 industries,such as manufacturing,retail,IT services,as well as state and federal governments,and featured a mix of B2B(51%)and B2C(49%).The study covered several key areas,including demographics,organizational performance,strategic priorities,data connectivity to Salesforce,AI adoption and expectations,as well as operational challenges.The survey also explored how companies manage change,AI/agentic AI investments,and make decisions.Additionally,it assessed leadership approaches,talent strategies,and cultural readiness for transformation,as well as collaboration efforts and regulatory concerns.The analysis of the 2025 Salesforce survey includes a robust mix of quantitative and analytical methods to uncover meaningful insights.Quantitative analysis begins with descriptive statisticssuch as means and frequency distributions to summarize key themes and technology adoption rates across the sample.To uncover the true drivers of performance,we created a composite index based on key capability areas that map to three themes(“Connect,”“Optimize,”and“Govern”).The underlying variables were selected through both theoretical and empirical justification,chosen for their alignment with core digital transformation constructs and their strong correlations with financial KPIs such as revenue growth and profitability.We then applied advanced analytics,including correlation analysis,to examine the relationship between these capabilities and business outcomes.To deepen our insights,we segmented the composite score into quintiles,which allowed us to explore how organizations in the top quintile differ from others and to identify the practices most strongly associated with high performance.Each quintile was additionally normalized by annual revenue size.US25%UK10%Japan 9%Australia5nada5%France4%Germany4%India4%South Korea4%Spain4%Singapore3%Switzerland3%UAE3%Argentina2%Brazil2%China(Hong Kong only)2nmark2%Mexico 2%Saudi Arabia2%Sweden2%Finland1%Netherlands1%Qatar1%Respondents by industryRespondents by country31NOTE:Percentages may not add to 100%due to rounding.Manufacturing/Industrial Products8%Consumer Products7%Telecommunications6%Energy and Utilities6%Retail6%Life Sciences/Pharmaceutical6%Healthcare Providers6%Travel5%Transportation/Logistics 5%Media and Entertainment5%Software 5nking4%Electronics4%Insurance4%Automotive OEM3%Automotive Suppliers3%Chemicals3%Financial Markets3%Government Federal/National3%Government State/Provincial3%Petroleum(inc.Oil and Gas)3%Professional Services32Notes and sources1 The State of Salesforce 2024-2025:3 priorities for driving an AI advantage.IBM Institute for Business Value.September 2024.https:/ 2025 CMO study.The CMO revolution:5 growth moves to win with AI.IBM Institute for Business Value.June 2025.https:/ibm.co/2025-cmo3 The State of Salesforce 2024-2025:3 priorities for driving an AI advantage.IBM Institute for Business Value.September 2024.https:/ Ziv,Lior.“The Top Agentic AI Security Threats You Need to Know in 2025.”Lasso blog.May 18,2025.https:/www.lasso.security/blog/agentic-ai-security-threats-20255“Revolutionizing financial inclusion:How NDS Cognitive Labs transforms customer engagement with gen AI.”IBM case study.Accessed August 19,2025.https:/ 2025 CMO study.The CMO revolution:5 growth moves to win with AI.IBM Institute for Business Value.June 2025.https:/ibm.co/2025-cmo7“Modernized design.Maximized care.Water Corporation implements an eConcessions platform with IBM Consulting.”IBM case study.Accessed August 19,2025.https:/ Copyright IBM Corporation 2025IBM Corporation New Orchard Road Armonk,NY 10504Produced in the United States of America|October 2025IBM,the IBM logo,and are trademarks of International Business Machines Corp.,registered in many jurisdictions worldwide.Other product and service names might be trademarks of IBM or other companies.A current list of IBM trademarks is available on the web at“Copyright and trademark information”at: document is current as of the initial date of publication and may be changed by IBM at any time.Not all offerings are available in every country in which IBM operates.THE INFORMATION IN THIS DOCUMENT IS PROVIDED“AS IS”WITHOUT ANY WARRANTY,EXPRESS OR IMPLIED,INCLUDING WITHOUT ANY WARRANTIES OF MERCHANTABILITY,FITNESS FOR A PARTICULAR PURPOSE AND ANY WARRANTY OR CONDITION OF NON-INFRINGEMENT.IBM products are warranted according to the terms and conditions of the agreements under which they are provided.This report is intended for general guidance only.It is not intended to be a substitute for detailed research or the exercise of professional judgment.IBM shall not be responsible for any loss whatsoever sustained by any organization or person who relies on this publication.Examples presented are illustrative only.Actual results will vary based on client configurations and conditions and,therefore,generally expected results cannot be provided.1443d5dcbc4f4cc0-USEN-0033
2025-10-29
36页
5星级
Produced in partnership withAs AI becomes central to the enterprise,data engineers are stepping out from behind the scenes to help shape AI strategy and influence business decisions.Redefining data engineering in the age of AI2 MIT Technology Review InsightsPreface“Redefining data engineering in the age of AI”is an MIT Technology Review Insights report sponsored by Snowflake.This report,based on survey research and executive interviews,seeks to understand how the role of data engineering teams is evolving as AI becomes central to enterprises and their success.Denis McCauley was the author of the report,Virginia Wilson was the editor,and Nicola Crepaldi was the publisher.The research is editorially independent,and the views expressed are those of MIT Technology Review Insights.We would like to thank the following executives for their time and insights:Chris Child,vice president of product,data engineering,Snowflake Ritu Jyoti,chief executive officer,stealth AI startup(formerly general manager and group vice president of AI,automation,and data and analytics,IDC)Dave Masino,senior director,data and intelligence,SlalomGeorge Westerman,senior lecturer and principal research scientist,MIT Sloan School of ManagementMethodologyIn June 2025,MIT Technology Review Insights,in collaboration with Snowflake,conducted a survey of 400 chief information officers,chief technology officers,chief data and analytics officers,and other senior data and technology executives.The survey respondents work in organizations that span seven industries,each headquartered in one of 10 countries.All organizations earn$500 million or more in annual revenue.In addition to the quantitative research from the survey,a series of in-depth interviews with senior technology executives and other experts offer firsthand insights into the evolution of the data engineering teams role in the AI era.3MIT Technology Review InsightsCONTENTS01 Executive summary.402 Role change:The AI impact.5 Growing workloads and a shifting focus.7 Learning from software engineers.903 The agentic challenge.11 Agents of change.12 Risk equation.1304 AI-powered data engineering.14 More tools,greater complexity.15 Focus on finance and manufacturing.1605 Building influence.17 Engineer and architect.1806 Conclusion:Embracing the change.194 MIT Technology Review Insights02020101As organizations weave AI into more of their operations,senior executives are realizing data engineers hold a central role in bringing these initiatives to life.After all,AI only delivers when you have large amounts of reliable and well-managed,high-quality data.Indeed,this report finds that data engineers play a pivotal role in their organizations as enablers of AI.And in so doing,they are integral to the overall success of the business.According to the results of a survey of 400 senior data and technology executives,conducted by MIT Technology Review Insights,data engineers have become influential in areas that extend well beyond their traditional remit as pipeline managers.The technology is also changing how data engineers work,with the balance of their time shifting from core data management tasks toward AI-specific activities.As their influence grows,so do the challenges data engineers face.A major one is dealing with greater complexity,as more advanced AI models elevate the importance of managing unstructured data and real-time pipelines.Another challenge is managing expanding workloads;data engineers are being asked to do more today than ever before,and thats not likely to change.4 MIT Technology Review InsightsExecutive summary Data engineers are integral to the business.This is the view of 72%of the surveyed technology leadersand 86%of those in the surveys biggest organizations,where AI maturity is greatest.It is a view held especially strongly among executives in financial services and manufacturing companies.AI is changing everything data engineers do.The share of time data engineers spend each day on AI projects has nearly doubled in the past two years,from an average of 19%in 2023 to 37%in 2025,according to our survey.Respondents expect this figure to continue rising to an average of 61%in two years time.This is also contributing to bigger data engineer workloads;most respondents(77%)see these growing increasingly heavy.AI boosts data engineers productivity.Nearly three quarters of the business leaders surveyed(74%)say AI has led to an increase in the output of data engineering teams in the past two years.And for 77%,it has brought an improvement in the quality of their teams work.AI agents and tools increase efficiency,but also complexity.Over half of the surveyed organizations(54%)expect to begin deploying agentic AI within the next year(20%have already done so).For data engineering teams,agentic AI and other AI-powered tools should further enhance efficiency by automating tasks in pipeline optimization,data integration,orchestration,and other areas.But respondents also worry that new AI tools will increase integration complexity and add to tech stack fragmentation.Data engineers influence extends beyond pipeline management.Two-thirds of respondents(66%)say their data engineers are influential in decisions about investment in data tools and on vendor selection.Over half say the same about overall data strategy(51%),AI use-case feasibility(53%),and business units use of AI models(56%).Key findings from the report include the following:5MIT Technology Review Insights0202Role change:The AI impactData engineers have worked behind the scenes for years but are now capturing the attention of their organizations senior leaders.This is largely thanks to AI.Of the senior technology leaders in our survey,72em data engineers to be integral to the overall success of their business(see Figure 2a).This figure rises to 86%of respondents in the surveys largest organizationsthose with annual revenues of more than$10 billion(see Figure 2b).Sector-wise,executives in financial services and manufacturing are particularly convinced of the importance of data engineers.According to Ritu Jyoti,formerly a senior executive at International Data Corporation(IDC)and now chief executive officer of a stealth AI startup,the explanation for data engineers elevated profile is straightforward:“AI is increasingly a foundation of business success,and there is no AI without data.”Chris Child,vice president of product,data engineering at Snowflake,makes a similar point:“To realize AIs full potential,a strong data foundation isnt optionalits mission critical.And were seeing the workforce evolve to follow suit,with data engineers being viewed less Figure 1a:Survey respondents by country(where their organization is headquartered)(%of respondents)Source:MIT Technology Review Insights survey,2025Figure 1b:Survey respondents by primary industry(%of respondents)Source:MIT Technology Review Insights survey,202535%9%8%8%5%5%5%5%5%USACanadaUKFranceGermanyAustraliaIndiaJapanNew ZealandSouth Korea18%Financial services(includingbanking&insurance)RetailConsumer packaged goodsManufacturingHealthcare&life sciencesMedia&entertainmentAdvertising6 MIT Technology Review Insightsas a backend plumbing function,but instead as an integral,strategic business partner responsible for enterprises most valuable asset:their data.If your C-suite still considers data engineering as a support role,youre already five years behindand probably training your future competitors.”For the surveys biggest companies,which placed even greater emphasis on the importance of data engineers,this is likely due to their more advanced AI maturity.Considerably more of the biggest companies have already deployed generative,multi-modal,and semi-autonomous agentic AI,as well as AI-powered data tools.“Some of those early AI adoptersthe organizations that are really moving the needle on AIare looking at their data engineers as strategic enablers in business transformation,”says Jyoti.“To realize AIs full potential,a strong data foundation isnt optionalits mission critical.If your C-suite still considersdata engineering as a support role,youre already five years behindand probably training your future competitors.”Chris Child,Vice President of Product,Data Engineering,Snowflake Figure 2b:Nearly 9 in 10 of the largest surveyed organizations say data engineers are integral to the success of their business Source:MIT Technology Review Insights survey,2025Figure 2a:Nearly 3 in 4 data and technology executives agree data engineers are integral to the success of their business(%of respondents)Source:MIT Technology Review Insights survey,202520#79D88HEQBD9IB%Strongly agreeSomewhat agreeMedia&entertainmentRetailCPGHealthcare&life sciencesAdvertisingManufacturingFinancial services$500 million to$1 billion$1 billion to$10 billionMore than$10 billionBy industryBy company size28D!%7%0%Strongly agreeSomewhat agreeNeutralSomewhat disagreeStrongly disagree7MIT Technology Review InsightsGrowing workloads and a shifting focusA large majority of surveyed executives(81%)say the job description for data engineers in their organization has changed radically due to AI.This can be seen in a shift in their day-to-day activities.Two years ago,data engineers spent an average of 19%of their workday on AI projects(such as building and monitoring model pipelines or data cleansing),say respondents,and 81%on other activities(data warehousing,quality checking,and infrastructure management,for example).In 2025,the analogous figures are 37%of time spent on AI and 63%on other activities(see Figure 3).“The role has evolved significantly,”says Jyoti.“In the earlier days of AI and machine learning,data engineers focused on building data pipelines.They were very siloed,focusing on one task at a time.Now theyre focused on designing and architecting reusable data platforms,governance,and oversight.Theyre playing a pivotal role as key enablers of AI.”The respondents expect the change to accelerate:two years from now,they believe the ratio of time spent on AI to that on other activities will almost be the reverse of today61%to 39%in favor of AI(see Figure 3).As priorities change,so too does the volume of work.Most respondents(77%)say their data engineers workload is becoming increasingly heavy(see Figure 4a).“Theyre struggling to keep up,”says Jyoti.“Data volumes and velocity have exploded.Data is now multi-modal and is being ingested in real time.”Interestingly,considerably more chief AI officers observe this than their peers(see Figure 4b).“Some of those early AI adoptersthe organizations that are really moving the needle on AIare looking at their data engineers as strategic enablers in business transformation.”Ritu Jyoti,Chief Executive Officer,Stealth AI StartupFigure 3:Data engineers are spending increasing amounts of time on AI projects versus other activities(%of respondents)Source:MIT Technology Review Insights survey,2025197ca9%Average time spent on AIAverage time spent on other activitiesTwo years agoTodayTwo years from now8 MIT Technology Review Insights“Models need so much more data and in multiple formats,”says George Westerman,senior lecturer and principal research scientist at the MIT Sloan School of Management.“Where it used to be making sense of structured data,which was relatively straightforward,now its:What do we do with all this unstructured data?How do we tag it?How do we organize it?How do we store it?Thats a bigger challenge.”Dave Masino,senior director of data and intelligence at technology consulting company Slalom,sees the growing workload as an opportunity.“AI has increased the amount of work data engineers are doing,but you inherently want your team to be busy,”he says.“I see the advantage were getting from AI-enabled acceleration eclipsing the amount of additional workload.”“Models need so much more data and in multiple formats.Where it used to be making sense of structured data,which was relatively straightforward,now its:What do we do with all this unstructured data?How do we tag it?How do we organize it?How do we store it?Thats a bigger challenge.”George Westerman,Senior Lecturer and Principal Research Scientist,MIT Sloan School of ManagementFigure 4a:3 in 4 respondents agree data engineers workloads are growing(%of respondents)34C %3%Strongly agreeSomewhat agreeNeutralSomewhat disagreeFigure 4b:CAIOs are the most likely to agree that data engineers workloads are becoming increasingly heavy(By executive role)Source:MIT Technology Review Insights survey,2025Strongly agreeSomewhat agree313D%872S5VS%Chief data/analytics ofcerChief technology ofcerSVP/VP/head of IT,data,AI,engineering,technology,or similarChief information ofcerChief experience ofcerChief enterprise/data architectChief AI ofcerSource:MIT Technology Review Insights survey,20259MIT Technology Review InsightsData engineers success in mastering AI can in large part be credited to their ability to adapt software engineering practices to their own needs,says Chris Child,vice president of product and data engineering at Snowflake.“To my mind this is the biggest shift in data engineering in the past five to seven years,”he says,“and its still ongoing.Its helped data engineers understand how to build and manage complex software projects,including AI projects,at scale.”Child provides examples:“Its how to treat infrastructure as code,using tools like Terraform to set up your data engineering infrastructure.Its making sure all your data engineering code is checked into version control deployed through continuous integration and continuous deployment.”This is helping with managing AI projects,Child says,because many AI tools in use today have been trained on software engineering problems.“Its been a massive positive,”says Child.“Data engineers are now thinking in a software kind of way.Theyre now better able to move quickly and to change things without fear of breaking something,”he says.“Its been a change in philosophy and team culture as much as a change in tooling.”Learning from software engineers“AI has increased the amount of work data engineers are doing,but you inherently want your team to be busy.I see the advantage were getting from AI-enabled acceleration eclipsing the amount of additional workload.”Dave Masino,Senior Director of Data and Intelligence,Slalom10 MIT Technology Review Insights0303Entering a golden age for data engineeringThe rise of AI has made data engineers more important than ever.With growing workloads and responsibilities,they are now strategic partners who directly shape business outcomes.At Snowflake,we see this as more than a simple shift in responsibilities.Its the dawn of a new,inspiring era for data engineering as a whole.Adapting to meet complex and growing demands is nothing new for data engineers;these days,theyre handling more unstructured data and managing real-time pipelines for AI.And this is where Snowflakes platform is designed to help them thrive.Our fully managed,serverless architecture handles the operational overhead,allowing data engineers the freedom to focus on strategic,high-value tasks instead of worrying about infrastructure provisioning,tuning,and scaling.We dont just provide the building blocks;we make it easy to create a resilient data foundation that is ready for the future.Delivering business value through trustworthy data without getting bogged down in mundane maintenance requires tools that simplify processes,and these are built directly into the Snowflake platform.Starting with ingestion,Snowflake Openflow makes it easy to handle multimodal data with high throughput and low latencywhich weve made even more cost-effective.With Snowpark,data engineers can use familiar languages like Python to build,manage,and deploy data pipelines and machine learning models,bringing the code to the data instead of the other way around.For real-time needs,Dynamic Tables offer a declarative framework that streamlines the creation of both batch and streaming pipelinesmeaning a single,consistent framework can work for both types of data flows without the traditional complexity.Given the shift in focus toward AI projects,as the survey revealed,we see the value in having a built-in suite of AI-powered capabilities,like Snowflake Cortex AI.By integrating data and AI on a single,secure platform,Snowflake can help data engineers move beyond basic data management and become the essential partners their businesses need to succeed with AI today.This,of course,also underscores the critical importance of data governance.New AI tools can often increase the complexity of integration and add to tech stack fragmentation,making robust governance more difficult to maintain.Snowflake is built to address this head-on.Our platforms unified architecture consolidates structured,unstructured,and semistructured data in a single,secure environment,minimizing the fragmentation that leads to gaps in governance.Providing features that help with secure data access,lineage tracing,and data qualityessential for maintaining the integrity of the data feeding AI modelsSnowflake can help data engineers trust their data.The role of the data engineer is expanding.Theyre architects,foundation-setters,orchestrators,and so much more.Simply put:They are the operational lifeblood of any data-driven organization.Partner perspective:Snowflake11MIT Technology Review Insights0303The agentic challengeThe relentless progression of AI means that data engineers must keep pace and grasp the data implications of the fields advancing capabilities.“Data engineers will need to keep increasing their efficiency,”says Child.“For one thing,the amount of data that AI models require is growing exponentially,as is the number of AI projects.”But its more than volume,he says.“Data engineers are managing more complexity,such as unstructured data and real-time pipelines.Theyre also managing the ever-increasing expectations of business stakeholders.”Data engineers in the surveyed organizations are clearly striving to meet those expectations,judging by how their workday has shifted since 2023(see Figure 3).That change has coincided with a surge in enterprise adoption of generative AI.Most of the surveyed businesses have by now deployed generative AI,and most of the rest will begin doing so in the next 12 months(see Figures 5 and 6).“Data engineers are managing more complexity,such as unstructured data and real-time pipelines.Theyre also managing the ever-increasing expectations of business stakeholders.”Chris Child,Vice President of Product,Data Engineering,SnowflakeFigure 5:More than 8 in 10 technology leaders say their organization has already deployed AI-based data engineering tools(%of respondents)Source:MIT Technology Review Insights survey,2025Adoption of multi-modal AI,which incorporates a wide variety of data types,is also growing.Four in ten of the surveyed businesses have already begun deploying multi-modal AI,while 31%plan to start doing so within the next year(see Figure 5).Multi-modal AI illustrates the imperative for data engineers to learn how to manage and optimize flows of unstructured and semi-structured data.The diversity of data typesimages,video,audio,sensor data,and other formatsis not new.But the volumes and varieties consumed by large language models(LLMs)and the speed of data ingestion are much greater than in pre-generative AI days.73 d!T11%Generative AIAgentic AIMulti-modal AIDomain-specific AmodelsILarge world modelsAI-based dataengineering toolsHave begun deployingWill begin deploying within 12 months from now12 MIT Technology Review InsightsAgentic AI with its autonomous capabilities poses the next big challenge to data engineers.Its adoption thus far is limited.Just one-fifth(20%)of the surveyed organizations have begun working with agentic models.Over half(54%)say they will start doing so in the next 12 months(see Figure 5).Agents of changeFor enterprises,the business benefits agentic AI offers are not vastly different from generative AI:gains in operational efficiency,employee productivity,and automation-driven cost economies,for example.The difference with agentic AI lies in its ability not only to analyze and inform decision-making as generative AI does,but to be goal-driven,adaptive,and autonomously make decisions and act on them.“Agentic AI relies on foundation models or LLMs as does generative AI,but it provides more advanced reasoning,context,interpretation,and self-correction,”says Jyoti.“Agentic AI will give us systems that not only research,analyze,and plan,but that act on plans in a dynamic and agile way.”For data engineering teams,agentic AI offers a raft of efficiency-enhancing benefits.The top two,according to surveyed respondents,are better pipeline debugging and optimization(cited by 42%)and better data integration(38%).The latter refers to the consolidation of data from different systems and formats to create single,consistent datasets.Improved orchestration across teams(34%)and stronger data governance and compliance(33%)are other frequently mentioned agentic AI advances for data engineering(see Figure 7).Among the surveys biggest businesses,improved orchestration is the most attractive potential benefit(cited by 52%).For the smallestthose with annual revenues of between$500 million and$1 billionit is data integration(47%).Child predicts higher value benefits coming from agentic AI.“Well start to see more agentic data engineering where having AI agents do a larger chunk of their operational work allows data engineers Figure 6:The majority of companies are already deploying generative AI.The largest are most advanced with agentic AI Source:MIT Technology Review Insights survey,2025“Agentic AI relies on foundation models or LLMs as does generative AI,but it provides more advanced reasoning,context,interpretation,and self-correction.Agentic AI will give us systems that not only research,analyze,and plan,but that act on plans in a dynamic and agile way.”Ritu Jyoti,Chief Executive Officer,Stealth AI Startup85ifxwriWC%80%4%#%More than$10 billion$1 billion to$10 billion$500 million to$1 billionFinancialservicesAdvertisingMedia&entertainmentHealthcare&life sciencesCPGRetailManufacturingGenerative AIAgentic AIBy company sizeBy industry13MIT Technology Review Insightsand teams to think about the bigger picture,”he says.“Theyll ask,What are our overarching goals?What budget do I give to which of these agents to process data?How do we think about our overall data estate rather than just individual pipelines?Then youll start to see a larger shift in the role of the data engineers.”“Im looking forward to offloading all the repetitive engineering work to AI agents and having my team focus on the things that are interesting,which are architecture,systems thinking,and solving real business problems,”says the head of data and analytics at a large retail organization.Risk equationWith advances in technologies like agentic AI also come challenges.For survey respondents,by far the most critical challenge is in ensuring data security and privacy,as cited by 55%(see Figure 8).The defining characteristics of agentic AI,as autonomous systems that take actions themselves,magnifies the potential fallout that could occur as a result of a data breach.“The best case scenario is that a breach results in some embarrassment.The worst case is that your business is forced to shut down,says Masino.Figure 7:Pipeline debugging and optimization and data integration are the top two benefits of agentic AI for data engineering teams (%of respondents)Source:MIT Technology Review Insights survey,2025Figure 8:Ensuring data security and privacy is rated the greatest challenge for data engineering teams as AI capabilities advance(%of respondents)Source:MIT Technology Review Insights survey,2025Pipeline debugging and optimizationData integrationOrchestration across teamsData governance and complianceData catalogingData transformationData cleansingData modeling and tuningInfrastructure provisioning4284322#U72(%!%Ensuring data security and privacyManaging real-time data pipelinesEnsuring quality of synthetic dataManaging growing volumes of unstructured dataReducing bias in dataKeeping individual engineers skills up to dateEnsuring data governanceUpdating data architecture to accommodate new capabilitiesTracing data lineage14 MIT Technology Review InsightsWhile data engineers are working to make AI more effective,AI is returning the favor.The technology has proven a boon to data engineer productivity,in terms of output quantity and quality.Nearly three-quarters of respondents(74%)report AI-led improvement in their data engineering teams productivity in the past two years in terms of quantity.This could include the number of outputs such as projects delivered and new code generated.Even more(77%)say the same about the quality of their data engineering teams work(data freshness,for example)(see Figure 9).One-fifth of the surveyed executives say the improvement in output quantity has been“significant,”and 29%say the same about the improvement in quality.0404AI-powered data engineeringThese improvements are no surprise to Masino.“With advancements in generative AI over the past two to three years and,most importantly,its integration into software development tooling,data engineers now have a very powerful accelerator at their disposal,”he says.The survey respondents expectations of pipeline optimization,data integration,orchestration,and governance gains from agentic AI(see Figure 7)indicate a strong belief that AI can deliver yet more productivity gains to data engineering teams.This is further supported by the finding that 83%of respondents say their companies have begun using AI-based data engineering tools(see Figure 5).Figure 9:AI has improved the quantity and quality of data engineering teams output in the last 2 years,according to 3 in 4 respondents(%of respondents)Source:MIT Technology Review Insights survey,202574w# %3%3%QuantityQualityImprovedRemain unchangedDeteriorated15MIT Technology Review InsightsMany such organizations are likely to be using AI code-generation tools of the type that appeared on the market earlier this decade.More recently,tools have emerged that use AI to automate or assist data engineers with data cleansing,integration,pipeline monitoring,metadata management,workflow orchestration,feature engineering,governance and compliance,and other tasks.“With the advanced tools we have now,its a lot easier to do things that once were very difficult,”says the head of data and analytics at a large retail organization.“Such tools are making data engineers much more efficient,”says Westerman.Masino agrees,believing that AI-enabled code editors and command line tools have had the most significant impact thus far.“My favorite tools today are Cursor,Claude Code,and Codex CLI,”says Masino.“What began as code completion and chat interfaces in development environments are now code agents that can act with autonomy.”More tools,greater complexityAI tools are not a panacea for data engineers,however.Many such tools run on LLMs,and some of the challenges AI tools present are not dissimilar to those in AI use cases elsewhere in the enterprise.When asked to list the main challenges for data engineers posed by AI tools,our survey respondents put integration complexity and data governance at the top(cited by 45%and 40%,respectively)(see Figure 10).Integration challenges arise from,for example,the need to pull together data from different,often legacy,tooling systems used by a team.Models underpinning AI engineering tools rely on clean,accurate,and high-quality data as much as other models do,thus requiring robust governance processes.For instance,AI data tools must be able to trace the lineage of data being ingested,to monitor for data drift,and to manage access control securely.The biggest companies in the survey rate data governance their greatest challenge.“With advancements in generative AI over the past two to three years and,most importantly,its integration into software development tooling,data engineers now have a very powerful accelerator at their disposal.”Dave Masino,Senior Director of Data and Intelligence,SlalomFigure 10:Integration complexity,data governance,and tool sprawl are the highest-rated challenges of new data tools,including those powered by AI(%of respondents)Source:MIT Technology Review Insights survey,2025Integration complexityData governance challengesTool sprawl and fragmentationUnanticipated costsLong learning curve to use new toolsKeeping up with change in toolsOrchestration complexity due to new toolsPotential for vendor lock-in4581$ MIT Technology Review InsightsCost and integration concerns are closely connected to another challenge high on the respondents list:tool sprawl and fragmentation(cited by 38%)(see Figure 10).Aside from direct costs(AI tools often do not come cheap),respondents are also likely worried about indirect costs,like the time lost to data engineering teams from having to maintain a growing array of often disconnected tools.To avoid the hazards of fragmentation,teams need to rigorously review and consolidate their existing stack of tools before investing in newer,AI-enabled ones.“Tooling is continuing to evolve,”says Masino.“There are myriad different choices,and its best to avoid hyper-analyzing all of them.They all have rough edges here and there,and my advice is pick one,start using it,form your opinion on it,and continue from there.”When dissecting the survey results by industry,financial services and manufacturing companies stand out from others in several aspects of the data engineering challenge.Take overall performance.More data engineering teams in these industries,as well as in media and entertainment,have registered productivity improvements in terms of both quantity and quality of output thanks to AI.It may be no coincidence that data engineers in manufacturing and financial services spend more time working on AI projects today than their peers in other industries(40%and 39%,respectively).And the same is expected to be true two years from now(65%and 64%,respectively)(see Figure 11).Their comparatively greater time spent with AI may help to explain a slightly greater prominence that data engineers in financial services and manufacturing enjoy in their organizations than their peers elsewhere.When asked about decision-making in a range of areas,executives from those industries are more likely to describe their data engineers as influential.This applies to areas such as investment in data tools,vendor selection,AI use-case feasibility,and overall data strategy.Perhaps a more telling manifestation of their influence is that more data engineers from those two industries,along with those from advertising,are deemed by respondents to be integral to the success of their business(Figure 2b).Focus on finance and manufacturingFigure 11:Data engineers in manufacturing and financial services spend more time,on average,on AI projects than other industries(By industry;%of respondents)Source:MIT Technology Review Insights survey,202519! 7967564aedaYWW%TotalManufacturingFinancialservicesHealthcare&life sciencesRetailCPGAdvertisingMedia&entertainmentTwo years agoTodayTwo years from now050517MIT Technology Review Insights0505Building influenceData engineers have long been described as the plumbers of the data worldtechnicians who maintain the pipelines critical to the smooth running of systems.In their traditional role as masters of extract,transform,load(ETL)work,data engineers have also been likened to bricklayers,says Westerman.“They may not seem as important as the people who design the building or who come to use it,but it cant be built without them,”he says.Limited to this role,data engineers would not be expected to exert significant influence with AI,data science,or other technology teams,to say nothing of the wider business.“If youre basically doing ETL work,you dont need a voice outside of your team,you just do what youre asked to do,”says Westerman.“If youre doing the architect job,on the other hand,then a voice becomes much more necessary.”In most organizations,the data architect has typically played the role of designer of data systems and infrastructure and definer of data standards.Our research points to some convergence of the data engineer and architect roles taking place in many organizations.In the survey,that is evidenced in the influence data engineers are seen to exert in areas that have historically been considered the domain of data architects.For example,51%of respondents say their data engineers influence their organizations overall data strategy to a greater or lesser extent.More say data engineers influence extends to investments in data technology and tools(66%),technology vendor selection(66%),and business units use of data and data tools(63%)(see Figure 12).In most organizations,data engineers influence also extends to the use of AI.Over half say this is the case with decisions made on AI use-case feasibility(53%)and on business units use of AI models(56%).Four in ten respondents(41%)say their data engineers voice is heard on questions regarding overall AI strategy(see Figure 12).“If youre basically doing extract,transform,load work,you dont need a voice outside of your team,you just do what youre asked to do.If youre doing the architect job,on the other hand,then a voice becomes much more necessary.”George Westerman,Senior Lecturer and Principal Research Scientist,MIT Sloan School of Management18 MIT Technology Review InsightsEngineer and architectNevertheless,these findings provide a clear indication of an expanding data engineer role.“Some data engineers are starting to think about business problems that need to be solved,”says Child.“Theyre asking,Where should we be investing resources?What should these different agents focus on?”“Over time,the data engineer role will shift from writing code for all pipelines toward managing the infrastructure that these are running in,orchestrating across a lot of these,and setting the rules and tests to make sure the right data is coming in,”says Child.Jyoti maintains that the data engineer and data architect roles will converge,and she sees that occurring now at a handful of organizations.“Data engineers need to become fluent in AI.That will help them grow into the roles of system or enterprise architects.”Jyoti sees enterprise architects playing an important role in driving the companys transformation.“They can proactively see who the players are in the industry and what the customers are demanding,”she says.“Theyll seek ways to innovate and to help prioritize the companys AI use cases.Theyll provide input on where the company should invest to ensure those use cases deliver a competitive advantage.That will definitely raise their profile among the senior business leaders.”There is something of a disconnect on these points,however,between senior technology leaders.In particular,chief information officers(CIOs)have a weaker perception of data engineer influence than chief data officers(CDOs)and chief AI officers(CAIOs).The gap between CIO and CDO/CAIO perceptions is especially wide when it comes to data engineers influence over matters of AI use.This suggests that data engineers value to the business is not fully appreciated across the senior leadership team;it certainly isnt by all CIOs.Supporting this view,far fewer CIOs(55%)than CDOs(80%)or CAIOs(82%)deem data engineers to be integral to the business.One plausible explanation for this divergence is the degree of visibility into data engineers day-to-day activities.CDOs and CAIOs simply have more of it than CIOs.“Its hard to place value on the things you cant see,”says Masino.“CDOs and CAIOs typically work more closely with data engineering teams than CIOs and know how difficult it can be to get value out of data.”The divergence in C-suite views can also be explained by the nature of the executives respective roles,according to Child.“Part of the CDOs and CAIOs job is to think about the future,while the CIO is thinking a lot more about the infrastructure.The latter is likely thinking about data engineers as part of the machine thats functioning but not about how its going to shift.That said,AI is pushing leaders to reconsider these processes.In a few years time,the share of CIOs agreeing that data engineers are integral to the business is likely to be significantly higher.”Figure 12:Data engineers are seen to exert decision-making influence in areas historically seen as the domain of data architects(%of respondents)Source:MIT Technology Review Insights survey,202516$0 9G6911%Business units use of AI modelsBusiness units use of data and data toolsTechnology vendor selectionInvestments in data technology and toolsDetermining feasibility of AI use casesOverall AI strategyOverall data strategyExtremely influential:their voice is decisiveSomewhat influential:their voice is important060619MIT Technology Review Insights0606Conclusion:Embracing the changeSome data engineers worry that AI will automate their jobs away.While AI may indeed automate away some of what they currently do,data engineering jobs are likely safe for the foreseeable future.After all,there will always be a need for problem-solving around data.AI is also giving data engineers a valuable opportunity to grow.Among other benefits,by taking on core data engineering tasks,AI-powered automation will lighten heavy data engineering workloads to allow a shift in focus from execution to strategy.“As AI agents take on more work,data engineers will spend less time on ETL and more on the data and AI strategizing that will help take the business forward,”says Child.For data engineers,three calls to action emerge from the research:Become an expert in AI.Data engineers need to understand how existing and emerging AI models,particularly LLMs,work and how they ingest,process,and verify data.Organizations should help their data engineers find and enroll in courses on,for example,machine learning frameworks,deep learning,feature engineering,and model evaluation.Heads of AI could also organize internal training in these areas.Understand the business.Data engineers need a strong understanding of key business objectives to be able to deliver value.They need to have conversations within different business units to better understand what they are trying to do and what data they need.Technology leaders can facilitate this by pushing the creation of cross-functional teams in which data engineers are temporarily embedded with product owners and other stakeholders in AI use cases.Build strong communication and presentation skills.As data engineers expand their influence,they need to be able to communicate in language that business units can understand.Data engineers who can strike a good balance between technical(including AI)acumen and the ability to collaborate,listen,and persuade will be most valuable to the business.“As AI agents take on more work,data engineers will spend less time on extract,transform,load work and more on the data and AI strategizing that will help take the business forward.”Chris Child,Vice President of Product,Data Engineering,Snowflake20 MIT Technology Review InsightsAbout MIT Technology Review InsightsMIT Technology Review Insights is the custom publishing division of MIT Technology Review,the worlds longest-running technology magazine,backed by the worlds foremost technology institutionproducing live events and research on the leading technology and business challenges of the day.Insights conducts qualitative and quantitative research and analysis in the US and abroad and publishes a wide variety of content,including articles,reports,infographics,videos,and podcasts.This content was researched,designed,and written entirely by human writers,editors,analysts,and illustrators.This includes the writing of surveys and collection of data for surveys.AI tools that may have been used were limited to secondary production processes that passed thorough human review.While every effort has been taken to verify the accuracy of this information,MIT Technology Review Insights cannot accept any responsibility or liability for reliance by any person in this report or any of the information,opinions,or conclusions set out in this report.Copyright MIT Technology Review Insights,2025.All rights reserved.About SnowflakeSnowflake is the platform for the AI era,making it easy for enterprises to innovate faster and get more value from data.More than 12,000 customers around the globe,including hundreds of the worlds largest companies,use Snowflakes AI Data Cloud to build,use,and share data,applications,and AI.With Snowflake,data and AI are transformative for everyone.Learn more at (NYSE:SNOW).IllustrationsIllustrations assembled by Tim Huxford with elements from Shutterstock and Adobe Stock.21MIT Technology Review InsightsMIT Technology Review I
2025-10-29
21页
5星级
Unlocking GBS Potential:The Transformative Power of AIDeloitte Agentic Global Business Services POV .
2025-10-28
27页
5星级
The Value of Open Source AI for APEC Economies A Review of Industry,Academic,and Open Source Evidenc.
2025-10-28
51页
5星级
Work,reworked:What it takes to win in the age of Agentic AIAgentic AI isnt just a powerful new tooli.
2025-10-28
14页
5星级
Artificial Intelligence Strategy U.S.Department of Energy i Section Title U.S.Department of Energy A.
2025-10-28
45页
5星级
TECH NAVIGATOR:SECOND EDITION APPLYING AGENTIC AI TO INDUSTRIESKnowledge Institute2|Tech Navigator|S.
2025-10-28
52页
5星级
How GenAI is Rewiring the DNA of JobsIndeeds AI at Work Report|2025The future of work and the role o.
2025-10-28
14页
5星级
Threat Modeling Revisited 2025 ISACA.All rights reserved.Information SecurityC O N T E NT S4Introduc.
2025-10-27
15页
5星级
Generative AI and foundation models in the EU:Uptake,opportunities,challenges,and a way forwardEurop.
2025-10-27
63页
5星级
Organizations overestimate their ransomware readiness in the face of an evolving ransomware landscap.
2025-10-27
21页
5星级
Issue BriefOctober 2025The Use of Open Models in ResearchAuthorsKyle MillerMia HoffmannRebecca Gelle.
2025-10-27
63页
5星级
IBM Institute for Business Value|Research InsightsSecure the post-quantum futureHow quantum safety s.
2025-10-27
44页
5星级
HOW AI WILL RESHAPE THE FRONTLINES OF UPSTREAM OIL&GASReasoning,Agentic Orchestration,and Multimodal.
2025-10-27
13页
5星级
Issue BriefSeptember 2025Harmonizing AI GuidanceDistilling Voluntary Standards and Best Practices in.
2025-10-27
79页
5星级
十五五规划建议全文（25页）.pdf
三个皮匠报告：2025银发经济生态：中国与全球实践白皮书（150页）.pdf
2025刘润年度演讲PPT：进化的力量.pdf
三个皮匠报告：2025中国情绪消费市场洞察报告（24页）.pdf
清华大学：2025年AIGC发展研究报告4.0版（152页）.pdf
三个皮匠报告：2025中国AI芯片市场洞察报告（24页）.pdf
麦肯锡：2025年人工智能发展态势报告：智能体、创新与转型（英文版）（32页）.pdf
三个皮匠报告：2025中国稀土产业市场洞察报告-从资源到战略武器，中美博弈的稀土战场（25页）.pdf
毕马威：2025年第四季度中国经济观察报告（82页）.pdf
Gartner：2026年十大战略技术趋势报告（28页）.pdf
0731-84720580
商务合作：really158d
友链申请 (QQ)：1737380874
最新报告
中英对照
全文搜索
报告精选
PDF上传翻译
多格式文档互转
入驻&报告售卖
会员权益
机构报告
券商研报
财报库
专题合集
英文报告
数据图表
会议报告
其他资源
研报
新质生产力
DeepSeek
低空经济
大模型
AI Agent
AI Infra
具身智能
自动驾驶
宠物
银发经济
人形机器人
企业出海
算力
微短剧
薪酬
白皮书
创新药
行业分析
个股研究
年报财报
IPO招股书
会议纪要
宏观策略
政策法规
其他
人工智能
信息科技
互联网
消费经济
汽车交通
电商零售
传媒娱乐
医疗健康
投资金融
能源环境
地产建筑
传统产业
英文报告
其它
行业聚焦
芯片产业
热点概念
全球咨询智库
人工智能
500强
新质生产力
会议峰会
新能源汽车
企业年报
互联网
公司研究
行业综观
消费教育
科技通信
医药健康
人力资源
投资金融
汽车产业
物流地产
电子商务
传统产业
传媒营销
其它
2025年养老经济/银发经济/长寿经济/银发族/老龄化报告合集（共50套打包）
2025年商业航天行业报告合集（共41套打包）
AI、科技与通信
广告、传媒与营销
消费、零售与支付
HR、文化与旅游
金融、保险与投资
能源、环境与工业
医疗制药与大健康
物流、地产与建筑
其他行业
AI ▪ 科技 ▪ 通信
数字化
金融财经
智能制造
电商传媒
地产建筑
医疗医学
能源化工
其他行业
收藏
下载
2025-11-19
AI查数
行业数据
政策法规
商业模式
产业链
竞争格局
市场规模
产业概述
其它
2025年
AI读财报
年报
一季报
半年报
三季报
IPO招股书
社会责任报告
A股
IPO申报
港股
美股&全球
新三板
微信扫码登录
手机快捷登录
账号登录
