印孚瑟斯：2022人工智能数据雷达-从数据科学到实际业务让人工智能成为现实（英文版）（36页）.pdf

《印孚瑟斯：2022人工智能数据雷达-从数据科学到实际业务让人工智能成为现实（英文版）（36页）.pdf》由会员分享，可在线阅读，更多相关《印孚瑟斯：2022人工智能数据雷达-从数据科学到实际业务让人工智能成为现实（英文版）（36页）.pdf（36页珍藏版）》请在三个皮匠报告文库上搜索。

1、01 11001110 01 11 00 1100 001 11 00 10 11 00 11 01 01 11001110 01 11 00 1100 001 11 00 10 11 00 11 01 1100 11 01 11001110 01 11 00 1100 001 11 00 10 11 00 11 01 DATA+AI RADAR 2022MAKING AI REAL:FROM DATA SCIENCE TO PRACTICAL BUSINESS2|DATA+AI RADAR 2022External Document 2022 Infosys LimitedAIContent

2、sDATA+AI RADAR 2022|3 External Document 2022 Infosys LimitedExecutive summary 5Companies are new to AI and it shows 6Most AI is basic,and not autonomous 7Humans limit the speed of AI 9Sparse satisfaction,but financial services a bright spot 10A conversation about AI with AI 12Data is not the new oil

3、 it needs a new metaphor 13Ever-evolving methods of data management 14Data sharing delivers a dual benefit 15Effective AI requires data to be fresh and clean 16Poor processes invite bias 18Advanced AI requires trust across all dimensions 19Trust is a blind spot 20The where,what,and who of AI 21Deep

4、learning helps achieve AI at scale 22Whats at stake:Data+AI must prove its nothing to fear 23Recommendations:Precision,trust,business value 241.Get your data right,and share it 252.Build trust in advanced AI 273.Compose an AI team biased to business value 28Appendix:Research approach 294|DATA+AI RAD

5、AR 2022External Document 2022 Infosys LimitedGenerating real value from data and AI requires companies to share data,build trust in the results delivered by advanced systems,and keep business goals at the fore.01 11001110 01 11 00 1100 001 11 00 10 11 00 11 01 1100 11 00 11 0001 11001110 01 11 00 11

6、00 001 11 00 10 11 00 11 01 1100 11 00 11 00 00101 11001110 01 11 00 1100 001 11 00 10 11 00 11 01 1100 11 00 11 0DATA+AI RADAR 2022|5 External Document 2022 Infosys LimitedMaking AI Real:From data science to practical businessExecutive summaryCompanies need to think differently about data and artif

7、icial intelligence(AI).They have invested heavily in AI systems in the past few years.Global spending on AI-centric systems will approach$118 billion in 2022 and grow to more than$300 billion by 2026.1 All that work is not paying dividends.Companies have achieved basic AI capabilities.This is not wh

8、at they want.Three out of four companies in our survey want to operate AI at enterprisescale.Infosys Knowledge Institutes inaugural Data+AI Radar identifies why AI fails to deliver on heightened expectations,and recommends three areas for improvement:develop data practices that encourage sharing,bin

9、d explanations into advanced AI,and focus AI teams on business.If companies improve on these fronts,they can add up to$467 billion in profit growth,collectively,and increase internal satisfaction with data and AI.Companies are new to advanced AI and it showsOur survey of 2,500 AI practitioners found

10、 that 81%deployed their first AI system in the past four years.However,most companies(85%)have not achieved advanced capabilities,and most AI models(63%)are still driven by humans.Compounding this,outcomes are middling at best:Users are highly satisfied with their data and AI results only about a qu

11、arter of the time.Data is not the new oilBusinesses can no longer afford to think of their data as oil,extracted with great effort and valuable only when refined.Data today is more like currency:It gains value when it circulates.This data-sharing economy is already up and running.In our study,compan

12、ies that shared data,in and out of their organization,are more likely to have higher revenue and use AI better.Refreshing data closer to real time also correlates with increased profits and revenue.However,companies cant share data until they trust it.Shortcomings in data verification,data practices

13、,and data strategies continue to hold companies back.Advanced AI requires trust in all directionsAdvanced AI requires a new way of thinking about technology and business.It requires trust:Trust in your own and others data management,and trust in AI models.Pristine data and perfectly programmed AI mo

14、dels mean nothing if humans do not trust and use what data and AI produce.Companies that are the most satisfied with their AI consistently have strong,responsible AI practices.Trust,share,and focus on valueOur analysis identifies ways data and AI work better together:1.Get your data right,and share

15、it.Focus on data sharing capabilities and hub-and-spoke data management.2.Build trust in advanced AI.Strengthen ethics,bias management,deep learning,AI cloud,and scaling across the organization.3.The AI team needs a bias for value.Business leaders matter as much as data scientists.Combined,these act

16、ions will scale AI and unlock value positively impacting the bottom line and user satisfaction.6|DATA+AI RADAR 2022External Document 2022 Infosys LimitedIts not hard to find a business that could be improved by data and AI.Simply look for an exercise with routine,highlymanual processes and lots of d

17、ata.That notion led many real estate companies to pursue“iBuying”developing algorithms to speed up and optimize residential home-buying in the US.Ithasntworked.Zillow,the market leader in public-facing residential real estate data,shut down its algorithm-driven iBuying unit in November 2021 and took

18、 a$300 million charge for the quarter.2 OpenDoor,a tech startup built on iBuying has yet to show a profit,in spite of raising$1.9 billion in eight years in business.3 Zillow began offering up computer-generated home estimates 15 years ago.After all that time,its algorithm still couldnt deliver relia

19、ble results.These outcomes should serve as cautionary tales,but scores of big businesses are coming new to AI,eager to apply data and AI in big ways.Figure 1.Most companies deployed their first AI system within the past four yearsCompanies are new to AI and it showsAs the saga of iBuying illustrates

20、,AI is not new.But its being newly applied in more businesses than ever.A decade ago,digital giants including Amazon,Google,and Microsoft accounted for most data and AI activity.4 Around the turn the millennium,these companies had two things going for them that most enterprises lacked:zettabytes of

21、data and petahertz of computing power.Now every company is like Amazon in 2001:Gallons of data flow into corporate reservoirs and computing power can be ramped up with the touch of a button.The result:AI is spreading like wildfire.But most companies are new to attempting their own advanced AI.We sur

22、veyed 2,500 AI practitioners from companies around the globe,and found that four out of five companies put their first AI model in production less than five years ago(Figure 1).To be sure,many enterprises have been making use of classic rule-based artificial AI and simple automation for years.For in

23、stance,a European paper producer has prioritized maintenance work and reduced pump failures in its paper mills using a system developed by Nokia,Infosys and the firm formerly known as Pyry(now a unit of AFRY).5But todays interest stems from the belief that AI can derive value from massive amounts of

24、 data and develop novel consumer products.6 Data scientists,AI experts,and academics have achieved significant advances in AI.7 But companies are struggling to answer:“What can AI do for business?”Most people,when they think of AI,think of super-advanced models designed to predict the future and cha

25、nge the business,says Balakrishna DR,Infosys senior vice president and head of AI and automation.“Companies want to use AI across their enterprise,to uncover hidden insights and construct new business models.But few are achieving this because it is a multi-faceted challenge involving data processes,

26、AI techniques and interdisciplinary teams,”he says.19%2017 or earlier30%2018-201938%202013%20211%2022Years of first AI deployment 81%DATA+AI RADAR 2022|7 External Document 2022 Infosys LimitedMore often than not,data and AI fail to deliver high satisfaction from users because most companies use only

27、 basic AI.A full 85%of AI practitioners have not achieved top-tier capabilities the sort of capabilities that are closer to AI that can predict the future.We asked respondents what capabilities AI systems deliver,and scored answers across our Sense,Understand,Respond,Evolve(SURE)taxonomy(Figure 2).T

28、he Infosys Knowledge Institute developed SURE with the guidance of AI expert Rajeshwari Ganesan.The framework draws inspiration from the eight layers of an AI stack articulated by Shailesh Kumar,chief data scientist at India telecom firm Jio.(He describes the eight-layer AI stack,or Ashtang-AI,in hi

29、s keynote address to the virtual 2020 Open Data Science conference.)8 The SURE taxonomy includes four tiers,ranging from the basic Sense capabilities(simple signal processing,such as being trained to recognize an object)to the most advanced,Evolve level(a system that senses,finds causes,acts on reco

30、mmendations,takes feedback,and refines its performance).Survey respondents said only 15%of AI systems reach the evolve tier.“The digital giants,including the cloud giants and others such as Apple,Facebook,and Netflix,are able to attain top-tier Evolve AI capabilities,but other large enterprises cant

31、,”explains Balakrishna.“The Fortune 100 arent there because of their old systems and ways.They want to do AI,but they dont know how to get something out of it.”More than one-third(36%;Figure 2)of AI delivers basic Sense capabilities.While that may not result in game-changing insights for an enterpri

32、se,its a pretty reliable use for AI,said John Bohannon,director of science at Primer.AI,a California-based AI and natural language processingbusiness.“From an engineering and data science point of view,monitoring is the most successful AI task,”he said.“Thats because the AI doesnt have to be as good

33、 as a human at it.The AI can provide value by prioritizing the flood of inbound information.”AI systems that rank and order enable workers to look at things more efficiently.Sensing inputs is also the most forgiving sort of AI when it comes to data quality.“You just have to match format,”Bohannon sa

34、ys.Sense-level AI typically comprises only a few different data formats.Of course,matching formats grows more complex as AI models grow more intricate.From a business perspective,these capabilities are commodities.Most companies have them,and they are not differentiators.“Companies need advanced AI

35、if they are to achieve the loftiest ambitions of AI and stand out from competitors,”says Sunil Senan,Infosys senior vice president head of data and analytics.Most AI is basic,and not autonomousFigure 2.Sense,Understand,Respond,Evolve(SURE)taxonomy:Only 15%achieve top AI capabilitiesCapabilityDefinit

36、ionExampleProportionRespond+train itself and improveDrug discoverysimulationsUnderstand+act autonomouslyAutomated loan decisioningSense+make predictionsForecast product demandIdentifypatternsEvolveRespondUnderstandSenseImage recognition15%22%27%36%37%advanced63%basic8|DATA+AI RADAR 2022External Docu

37、ment 2022 Infosys LimitedDATA+AI RADAR 2022|9 External Document 2022 Infosys LimitedThe SURE taxonomy reflects that companies cannot ascend to advanced capabilities without first achieving the basics.Think of AI in the context of an automobile.Collision-avoidance alarms and lane-assist technologies

38、are at the basic Sense and Understand levels.Adaptive cruise control and fully autonomous vehicles are at the advanced Respond and Evolve levels.Simply put,the big break between basic and advanced AI capabilities is whos in the drivers seat.For example,UK medical insurance company Bupa Global takes

39、in claims from 220 countries around the world,and has worked with automation and AI to extract relevant details from different sorts of claims submissions.“What youre saying to an AI is,Here is an invoice in a format youve never seen before.Can you pull the data out?”says Bupa Global IT strategy hea

40、d Steve Williams.“If you dont get to a high enough confidence level,someones still going to look at that piece of paper.”Williams said its important for companies to think about what problem they are trying to solve with AI.While using AI to review invoices globally still needs a human touch,he says

41、 Bupa is achieving good outputs with that process trained on UK health claims,where a monolithic National Health Service and standardize invoices make the AI learning more manageable.9 In our survey some 63%of AI models fall into the Sense and Understand capability levels,which require user interven

42、tions.This restricts ultrafast computers to run only at the speed of humans,and limits how much data and AI can do.This limitation is an artifact of early AI implementations in business,Rajeshwari notes.A decade ago,organizations typically began their AI work with rule-based automation Humans limit

43、the speed of AIusing tools such as robotic process automation(RPA).Once deployed,the rule-based system added intelligence incrementally.For example,RPA tools added process discovery using deep neural networks,or added intelligent document extraction using machine vision algorithms.In these systems,a

44、utomated AI is a small part embedded in a larger system.The challenge,though,is that when 63%of AI systems rely on humans,they cant be ported from automated to autonomous.Automated AI can be built incrementally from rule-based systems autonomous AI cannot.To operate in dynamic and unpredictable envi

45、ronments,autonomous AI systems must be constructed through reinforcement learning,generative adversarial networks(GANS),and advanced neural networks.All this enables companies to engage in continuous learning and evolve in response to new stimuli.The transition to a live enterprise that can sense,re

46、spond and evolve smoothly can be daunting.10 But this can be made easier by employing a technology architecture built with space and flexibility to incorporate the autonomous flow of information.Harnessing autonomous AI is not simply a matter of adding bandwidth and processing more bits.It requires

47、companies to think differently.“Autonomy is the way of the future,but I dont think its going to be a flip-the-switch situation where everything goes autonomous.Its going to be a slow,gradual change.”says Saurav Agarwal,founder and CEO of Siera.AI,an Austin,Texas,maker of safety and autonomous drivin

48、g systems for use in warehouses.“Were at that stage where people are trying to figure out,What can I do with the automation?Where can I plug it in?What can I do with it?”10|DATA+AI RADAR 2022External Document 2022 Infosys LimitedWith companies struggling with what their AI capabilities actually are(

49、basic)and what they hope they will achieve(advanced and autonomous),business and AI leaders must do more to manage expectations.More often than not,data and AI fail to deliver.We asked AI practitioners what they used their data and models for,and if it was working.Respondents rated satisfaction with

50、 five use cases for their industry.They said data and AI left them highly satisfied one out of four times(Figure 3).We also mapped use cases by satisfaction levels and usage levels(frequency of use case),and found that only 18 of 63(29%)scored in the high satisfaction zone(top of Figure 4).Additiona

51、lly,12 of the 18 high-satisfaction use cases are more basic detect or automate functions,such as enabling equipment to self-diagnose potential problems in high tech and manufacturing applications or automated compliance and customer services tools in financial services.The financial services industr

52、y recorded the strongest satisfaction with its data and AI uses.It is the only industry to rate all five of its use cases in the high-satisfaction category.As the financial services sector shifted from analog to digital,the quantity and complexity of products increased drastically,notes Mohit Joshi,

53、Infosys president and financial services technology expert.This creates new business opportunities that require robust data governance.Sparse satisfaction,but financial services a bright spotEnterprises were attracted to AI that is intelligent,fast,and automated.So far,most have built AI that is sim

54、ple,slow,and manual.“Companies must be able to securely and instantly share data across platforms and services to enable seamless services to the customer.A strong governance strategy for managing data while ensuring quality and minimizing risk will enable faster development and more sophisticated d

55、ata-driven decision-making capabilities,”he says.“A good mix of secure data controlled with the right privacy controls is a good way for firms to keep in line with regulation but still derive maximum value from the data for the end user,”he says.The next-highest rates of satisfaction are,in order,in

56、 retail and hospitality,healthcare,and high tech(Figure5).But general satisfaction doesnt mean companies have optimized AI for business.For example,in retail(Figure 6),checking for inventory and streamlining checkout received high satisfaction ratings.But flashier and popular tech,such as virtual re

57、ality and augmented reality,doesnt meet expectations.On the dissatisfied end of the spectrum,AI tools that aim to deliver personalized recommendations or products received lower satisfaction ratings.Telecom,energy,and CPG report low satisfaction with their data and AI uses.Financial services firms c

58、an unlock additional value for their customers by implementing strong privacy controls coupled with seamless data exchange with partners,amplifying the power of customers own data.”Mohit JoshiPresident,InfosysDATA+AI RADAR 2022|11 External Document 2022 Infosys LimitedUsage Satisfaction rate HigherN

59、eutralLower18of 63 usecases havehighusage,highersatisfactionSatisfaction and usage rates for the 12 industriesIndustry-specific use casesFinancial servicesRetail or hospitalityHealthcareHigh techManufacturingInsuranceLife sciencesAutomotiveLogistics or supply chainConsumer packaged goods(CPG)Energy,

60、mining,or utilitiesTelecommunicationsHigherNeutralSatisfactionLowerHigherNeutralLowerOptimized product availabilityDeliver productinformation ondemandStreamlined checkout experienceDemand planningPerform digital marketing analysisCreate AR and VR product trialsSatisfaction Usage Satisfaction and usa

61、ge levels for retailOverall satisfaction rates for data and AILower25%Higher26%Neutral49%Figure 3.Nothing special:AI outputs fail toexciteFigure 4.Only one in three highly satisfied with industry use casesFigure 5.Financial services most satisfied,telecom leastFigure 6.Retail:Low satisfaction for AI

62、 in AR and VR12|DATA+AI RADAR 2022External Document 2022 Infosys LimitedIn researching data and AI,the Knowledge Institute interviewed more than 100 AI,IT and business experts from Infosys,industry and academia.In the final stages of drafting Data+AI Radar we spoke with Infosys Consultings Chief AI

63、Architect Harald Gunia and Chief Data Scientist Serge Plata on some follow-up questions.In addition to their own expertise,they enlisted GPT-3,a popular AI language model that can generate its own answers,to weigh in on our questions.Heres how that conversation went.Knowledge Institute:Does the natu

64、re of some industries ask more from AI than others?GPT-3:There is no one-size-fits-all answer to this question,as the nature of AI and its applications vary greatly from industry to industry.However,some industries may require more AI resources than others,depending on the complexity of the tasks in

65、volved and the amount of data available.Harald Gunia:Thats generally true,but further industry-specific criteria are:type of data(life science&insurance have more texts,images&time series and hence can make more use of deep learning than consumer packaged goods and banking),type of regulation(the bl

66、ack-box nature of machine learning makes its use harder in tightly regulated industries like life science&banking),and importance of mobility(industries like agribusiness and logistics require AI-solutions to be integrated into 5G-frameworks).Knowledge Institute:How can you keep the AI effort focuse

67、d on the most critical business problems?GPT-3:There are a few ways to keep the AI effort focused on the most critical business problems:1.Define the business problem that you want to solve with AI,and then identify the AI technology or approach that is best suited to solving that problem.2.Work wit

68、h business stakeholders to identify the most critical business problems that AI can help with,and then prioritize those problems.3.Conduct a feasibility study to assess whether AI can help solve a particular business problem,and if so,how effective it is likely to be.Harald:I would also recommend us

69、ing AI-specific business case templates with successfully calculated example cases.Great candidates of critical business problems in the past dealt with time-critical decisions to be taken based on incomplete information,processing information-rich unstructured data or acceleration/automation of rou

70、tine tasks,where humans have become unavailable or too expensiveSerge Plata:You need to superimpose a continuous improvement program on the AI roadmap.the secret is to align the AI resources and data strategy with the overall strategy of the business.Knowledge Institute:Is there something specific t

71、o telco,energy or CPG that makes good AI uses more challenging?GPT-3:There is no one-size-fits-all answer to this question,as the challenges associated with AI applications in telco,energy and CPG industries vary depending on the specific use case.However,some common challenges include the need for

72、large amounts of data to train AI models,and the need for specialized expertise to develop and deploy AI applications.Harald:More specifically,telco and energy have a lot of time-series data,which require specialized machine learning models.All industries have data-privacy challenges,while computer

73、vision plays an increasing role in CPG(in the form of visual search and visual question-answering).Serge:Agree with Harald.Data privacy is a main issue in all industries,but especially in telco.Another aspect that concerns many energy companies is social responsibility,so AI should also be responsib

74、le AI.In line with our survey results,GPT-3 delivers agreeable but not particularly satisfying answers to our questions.Cocktail party-grade insights about emerging technology are not the outputs companies want to achieve from their data and AI.A conversation about AI with AI DATA+AI RADAR 2022|13 E

75、xternal Document 2022 Infosys LimitedData is not the new oil it needs a new metaphorSixteen years ago,business strategists began talking about data as the new oil.11 Data,like oil,is hard to extract and valuable only after it is refined,they argued.And for many global enterprises in 2006,good,refine

76、d data was rare and held a potential akin to jet fuel.Today,acquiring and manipulating data is fairly frictionless.Drilling for oil remains a complex engineering feat.Refining crude oil takes enormous capital expenditures,specialized equipment,and trained professionals.Refining data still comes with

77、 challenges,but can be done with on-demand cloud,off-the-shelf tools,and citizen data scientists.In 2022,data needs a better metaphor.For organizations trying to extract value from data,we believe there are three more modern metaphors that can help them keep in mind the challenges and best practices

78、 they require for success:Data is more like nuclear power.Data is enriched with potential,in need of special handling,and dangerous if you lose control.The advance to data in cloud and AI in cloud have flipped the script for data.Companies dont lose data.Twenty-first century data has a long half-lif

79、e.When to use it,where to use it,and how to control it are as critical as where to put it.Business leaders see a full spectrum of AI-informed data uses,Suresh Renganathan,chief technology officer at Teachers Federal Credit Union says.“All leaders would like to see a data-driven business model.Datas

80、a fuel to propel business growth right now,”he says.“We want to accomplish personalized experience.We want to enable fast investment decisions.We want to predict and reduce delinquencies,and to track branch performances against objectives.There is a plethora of usecases.”Data is a new currency.It ga

81、ins value when it circulates.Companies that import data and share their own data more extensively achieve better financial results and show greater progress toward operating AI at enterprise scale a critical goal for three out of four companies in our survey.Companies recognize that the emerging dat

82、a economy holds great potential.But as recently as last year,only about one-third had taken steps to start collaborating with partners.12 Building the capability and the will to share data between customers and suppliers could drive tremendous value in the manufacturing space,says Priya Almelkar,vic

83、e president of IT manufacturing operations at North Carolina semiconductor firm Wolfspeed.“Data becomes the new currency because thats how youre able to add more revenue from those shared insights,”she says.“Its a new culture for semiconductor makers to get to the point where they feel comfortable s

84、haring their data across.From a manufacturing background,that has been a big no-no,I think now companies are opening up to be willing to share data.”Data is gold yet to be mined.Data is a mountain of material.Below the topsoil and above the bedrock is a mix of minerals varying in value.The challenge

85、 for the data and AI team is to locate and mine the gold,and leave the lead.Like the unmined reservoir,data doesnt have any intrinsic value,until you know whats in it,says Sameli Menp,chief data officer of OP Financial in Finland.“When you know whats in it,you can derive assets out of it,”he says.“I

86、t only has value if you can connect it to something real.”14|DATA+AI RADAR 2022External Document 2022 Infosys LimitedCompanies need a clear and comprehensive data strategy to manage data properly and ingest new data smoothly.The trouble is,most companies dont have a consistent data management strate

87、gy.Respondents tell us they want to manage data centrally,but this is not what most do right now.Our analysis shows that centralized data management links to better profit and revenue growth.However,a shift to fully federated data management also increases profit growth.Figure 7.Companies moving to

88、centralized data managementIf companies(and their technology partners)can execute,centralized data management will be the most common strategy for big businesses in 2023(Figure 7).But this doesnt necessarily mean its the best choice.In fact,our analysis found that most AI practitioners who currently

89、 have centralized data management intend to change strategies,with the biggest portion shifting to federated approaches.Figure 8 shows this flip-flop from centralized to federated,and vice versa.Most organizations have yet to settle on their preferred strategy,and in this immature phase,the market i

90、s yo-yo-ing between the two extremes.The reality is that these two extremes are too simple to adequately serve as a comprehensive corporate data strategy.Companies deal with so much data in so many sources for so many uses,a one-size-fits-all solution does not work.And yet a data warehouse,data esta

91、te,data lake,or data lakehouse without a central organizational authority would leave an organization at risk of a metaphorical data meltdown.Over time we expect more organizations to find a sophisticated balance of centralization and federation a middle ground that suits their context and needs.Raj

92、 Savoor,AT&T Labs vice president of network analysts and automation,shares the story of how AT&T organized its big data and then made it accessible for AI uses.The US telecommunications giant first invested heavily in big data management capabilities,and its chief data officer put extra effort in es

93、tablishing a democratized ecosystem where data and AI capabilities can be put to work,he says.“Theres a step function here in complexity as the amount of data increases we get a kind of finer grain visibility and we have a lot more intelligent controls to then apply decisions,”Savoor told Laurel Rum

94、a with MIT Technology Reviews Business Lab podcast.13 This structure has allowed AT&T to cultivate many AI use cases ranging from internal planning to customer support and threat detection.It also has led the company to better use feedback loops to optimize its models and develop additional use case

95、s,Savoor says.Figure 8.Strategy in flux:No clear pattern emergesEver-evolving methods of data managementFederatedMixedPresentCentralizedNext year39%32%19%34%49%26%Centralized,26%Centralized,49%Mixed,34%Mixed,19%Data management strategyFederated,39%Federated,32%PresentNext yearDATA+AI RADAR 2022|15 E

96、xternal Document 2022 Infosys LimitedData sharing delivers a dual benefitData in the 21st century is not a scarce,nonrenewable asset.Savvy businesses know that establishing a data-sharing ecosystem with partners and peers delivers greater benefits than a solitary data lake or warehouse.Siera.AIs Aga

97、rwal says his customers have grown to understand that sharing data with Siera benefits the borrower and thelender.“Any company that needs to build a good AI needs data for the system to train and get better.Customers these days are asking us,What can you do with our data?Can you give us custom AI mo

98、dels?”he says.Sieras best and toughest customers are the ones who push the company to engage in the most vigorous data sharing.“They give you access to their people,they give you access to their resources,they give you access to data,they give you feedback to give you insights.”“Weve had a customer

99、who has come down to our facilities and spent four days with us,testing the equipment and giving us different test cases and scenarios,throwing curveballs at us.The ones that help us make the technology better,they are the early adopters.”Inbound data sharing and outbound data sharing give companies

100、 new ways to ensure they have the right information for their data scientists and AI models.All else being equal,importing data from third parties and high levels of data sharing delivered bigger boosts to the corporate bottom line than any other data or AI action.Of the$467 billion in global profit

101、 increase available,$105 billion of that links to importing 75%or more of data from third parties,our analysis shows.Companies that have built a foundation to trust and share their own data can be more agile in their AI work and using AI at scale more readily,says Satish H.C.,Infosys executive vice

102、president and co-head delivery.Businesses that lack that cant clearly see the value of their data or trust it.Companies that dont trust their data risk getting caught in a vicious cycle of experimenting and side work preparing and cleaning data for a specific task.“Then youre working very hard and o

103、nly able to use data and AI to solve small problems,”says Satish.16|DATA+AI RADAR 2022External Document 2022 Infosys LimitedData and AI work best with fresh data.Companies that refresh the data in their AI models in near real time or at minimum every six months after launching a model achieve better

104、 financial outcomes.Our respondents are doing well on this front.Some 79%of respondents refresh data in real time or less than every six months.But for all that progress on keeping data fresh,companies do not have clean data Data verification was one of the top challenges facing analytics and AI tea

105、ms,along with AI infrastructure and compute resources(Figure 9).In other words,data scientists are needed to do as much data cleaning as they do science.Jacqueline De Rojas,president of techUK,a national trade group said that low-quality data emitted from legacy systems is the most hidden and import

106、ant challenge for driving AI adoption.“The power of algorithms has to be driven on clean data,”says de Rojas.14Northwestern University Professor and AI expert Mohanbir Sawhney said AI in business is at a critical juncture.In the way e-commerce and commerce were once two separate things that inevitab

107、ly united into an omnichannel,data analysis has entered AI modelings field of gravity.And yet,the challenge of data verification persists,regardless of experience(Figure 10)and even though cloud-based AI systems have made AI computing power more available and less costly.Other major challenges,such

108、as bias management and framing AI questions,can be managed with experience.Time and experience also improve companies abilities to manage biases.AI practitioners at companies that have been implementing AI for more than five years identify bias management as less of a problem than companies that are

109、 newer to AI.Figure 9.Data verification and AI infrastructure are top challengesEffective AI requires data to be fresh and clean15%15%13%13%10%Data verificationAI infrastructure andcompute resourcesRisk of bias in AIClearly identifying theproblems for AI to addressInsufficient subjectmatter knowledg

110、ePercentage of respondents ranking an item as a top three challenge,weighted by number of AI systemsDATA+AI RADAR 2022|17 External Document 2022 Infosys LimitedFigure 10.Data verification is a persistently high concernExploratory data analysis and data cleaning take time.Companies cant build a model

111、 until they know and trust that data is clean and accurate.Its common to spend two months exploring and cleaning data for every one month experimenting with the model,our experts tell us.“Thirty-40%of the time,and 70%of the effort,is in data discovery,preparation,and augmentation,”says Karthik Andhi

112、yur Nagarajan,industry principal and data expert atInfosys.Data and AI return value only once a model is in the field and doing work but the model must use clean data.This has companies more focused on data governance,says Andrew Duncan,CEO and managing partner at Infosys Consulting.“Corporate clien

113、ts view the ability to collect,organize and analyze clean data as a new differentiator,”explains Duncan.“Companies must take a forensic approach to real-time data capture across all forms and formats both internally and externally since this will form the backbone of all business units.”Data verific

114、ationHigh concernLow concern5 yearsYears since first AI deploymentAI infrastructure and compute resourcesInsufficient subject matter knowledgeRisk of bias in AIClearly identifying the problems for AI to address“To be more democratic about data and AI decisions,you must be more dictatorial about your

115、 data.”Mohanbir SawhneyMcCormick Foundation Chair of Technology,Kellogg School of Management,Northwestern University18|DATA+AI RADAR 2022External Document 2022 Infosys LimitedPoor processes invite biasEven with clean data,AI practitioners face additional pitfalls,starting with missing data an inevit

116、able problem in most data sets.We used a question about how practitioners handle missing data as a proxy to assess data processes.We found that 34%of our survey respondents say they handle missing data in ways that can allow bias to creep into AI models(Figure 11).That is,they simply impute values f

117、or missing data based on other values in the data set or they delete observations.Both of these methods can introduce bias.Most commonly(42%of respondents),companies deal with missing data by studying the data set to see how much the unknowns may vary from the included data.This requires a high degr

118、ee of clarity in what the missing data ought to represent,says Rajeev Nayar,Infosys vice president and data strategist.“Anytime you introduce artifacts,theres some kind of problem that you introduce how do you minimize it?”Nayarsays.Advanced AI brings additional tools,such as generative adversarial

119、networks(GANS)and neural networks,that can shift this work from a slow,manual data science task to an exercise in AI itself.Adding programming logic to address edge cases is less popular,but holds more promise,than heavy data study and scrubbing,says Gary Bhattacharjee,Infosys data strategist.At tim

120、es,getting data processes right is something companies cannot control,because of regulations or business considerations.AI practitioners from the highly regulated healthcare and financial services industries were two to three times more likely to simply delete data from data sets as a way to manage

121、incomplete data.Because they deal in sensitive personal information with heavy regulations,many companies in those industries dont want to leverage sensitive data in AI models or for alternative uses,our experts tell us.Figure 11.One in three companies use data processes that increase the risk of AI

122、 bias42%23%26%8%Determine if there is a fundamental differencebetween missing and included data.If yes,create new variables.If no,drop the missing data.Add programming logic to the model to addressedge cases involving the missing dataImpute the data using other values in the data setRemove rows that

123、 contain blanks34%risk biaseddataDATA+AI RADAR 2022|19 External Document 2022 Infosys LimitedGrocery businesses operate on thin margins.So when a US grocer had a chance to save more than$4 million annually by automating work schedules,a well-defined process,it looked like an obvious move.The trouble

124、 was,automating this process would have shifted scheduling work shifts from human hands to a machine learning system.The leader of the business unit was against it,and employee union rules didnt immediately align with the software tools that the AI would use to put it in place.In the end,the grocer

125、never used the system,and scheduling remained a 40-hour-a-week job conducted by humans.The best data and the most elegant AI model amount to nothing if humans do not believe the system is effective,fair,and adaptable.Otherwise,AI systems end up like the grocery store scheduler:developed but unused.T

126、he most valuable and most used AI systems instill trust as they operate,Bohannon says.“When you get it right,youll find that explainability and trust is baked in,”he says.Our survey results support this.Companies that develop strong ethics(Figure 12,top)and bias management capabilities(Figure 12,bot

127、tom)report higher satisfaction and trust for their data and AI use cases.We measured AI ethics and bias management capabilities in 12 ways.Companies with high confidence in these measures are more likely to be satisfied with data and AI.This holds for every measure of ethics and bias control that we

128、 studied.Enterprises that have only recently deployed their first AI models can start by cataloging the usage of AI across the business,said Adriano Koshiyama,a research fellow in computer science at University College London and co-founder of Holistic AI Inc.,a London-based AI risk management firm.

129、“A simple strategy is to peg the cataloging system to other widely used and adopted processes,such as privacy risk assessments or information security reviews,”he says.“The next step is to start risk-assessing in a way that a priority list can emerge on what the focus of the enterprise should be.”Ov

130、er time,robust AI risk management practices will be a substantial component of new enterprise risk management processes,Koshiyama says.Figure 12.Strong ethics and bias management correlate with greater trust in AIAdvanced AI requires trust across all dimensionsFive measures of bias management1.Sampl

131、ing bias2.Participation bias3.Convergence bias4.Measurement bias5.Bias derived from overfitting the modelSeven measures of AI ethics1.Understandable models2.Clear,useful outputs3.Explainable algorithms4.Processes design to systematically detect bias5.Active efforts for humans to report bias6.Clear d

132、ata provenance7.Data stewardship including routine ethics,compliance and privacy reviews13%28%60%63%74%94%13%Very low26%Low42%Fair68%High77%HigherHigher satisfactionEthics confdenceBias management confdencePercentage reporting higher satisfaction with AIPercentage reporting higher satisfaction with

133、AIVery lowLowerLowFairHighHigherEthics confidence and AI satisfactionBias management confidence and AI satisfaction20|DATA+AI RADAR 2022External Document 2022 Infosys LimitedTrust is a blind spotTrust is important for simple automation-grade AI,and even more critical for advanced AI,where calculatio

134、ns are too fast or complex for humans to quickly understand.Primer.AIs Bohannon calls this explainability.“To get an AI system to the level of performance where you even have a shot at them performing well enough to be trusted,you have systems that help you understand,”he says.“Theres an explainabil

135、ity all built in around it.Thats the most sophisticated version of an AI system.Its very hard to achieve and wont work for all tasks today.”Where will it work?Bohannan says,in cases where AI practitioners have achieved very high data quality that is properly formatted,large in volume,sufficiently di

136、verse,and specifically instructive for the problem AI is aiming to address.This final“instructiveness”dimension of data quality is the most difficult to attain,and requires business knowledge along with AI and data science acumen,hesays.Autonomous AI and self-training AI are at the core of the corpo

137、rate dream of enterprise-wide AI systems.Nearly three out of every four AI practitioners surveyed want to scale AI across their enterprise.But only 7%say trust is a top challenge in scaling AI.Instead,they most frequently rated the business reality of managing costs(21%)as the top challenge(Figure 1

138、3).With economic uncertainty looming15,controlling costs are likely to continue to remain atop the list of corporate concerns.The challenge for companies focused on cost is to not neglect trust as they engage more with AI.Trust is the next big horizon in implementing AI systems,says Bonnie Holub,a d

139、ata science leader with Infosys Consulting.Teams have been struggling to simply get AI systems implemented at all,let alone efficiently,but best-in-breed companies are industrializing AI systems at scale now,shesays.“At that point,trust and responsible AI systems become a major issue.We see trust an

140、d responsible AI as crucial parts of the non-financial governance issues investors are demanding from companies,”Holub says.Figure 13.Despite its importance,trust rated a low concern by executives21%18%16%15%13%11%7%High costsDataAdaptabilityAI talentPartnershipsProblem-framingTrustPercentage of res

141、pondents that list a challenge in the top 2What are your companys top 2 challenges in scaling AI?DATA+AI RADAR 2022|21 External Document 2022 Infosys LimitedBefore businesses can scale AI,they must make some smart choices about deploying it.Our analysis shows that focusing AI on customer experience

142、and core operations can contribute to profit and revenue growth.In capability terms,the middle tiers(Understand and Respond)link to profit increases,while the most basic tier(Sense)and most advanced tier(Evolve)do not at this point deliver businessvalue.Adding systems at the Understand and Respond l

143、evels has a smaller impact on business outcomes than data-sharing behaviors.But these middle tiers represent a pivotal point in the maturity of an enterprises AI efforts.Siera.AI offers two types of autonomous AI systems for use with forklifts and industrial environments.On the basic end are tools t

144、hat prevent collisions.The higher end includes systems that study how things get moved and looks for ways to optimize warehouse operations.Agarwal says company managers get excited about the optimization and analysis.But then they hear a different story from the workers closest to the systems.“The p

145、eople on the ground level feel like theres a computer watching them and theres a tattletale,”he says.“I realized that what they really wanted was something between the two.”Perhaps that system could eventually add more analysis and autonomy down the road,but more immediately warehouse managers and g

146、round-level employees need tools that enhance safety and make their people feel comfortable adopting it.“I think thats been the biggest,aha,moment.This is not about the sexiest technology that the people at the top want.Theres a middle ground,where technology is advanced,but simple enough so that pe

147、ople at the top and the bottom can meet in the middle,”Agarwal says.Holub,editor of Infosys Consultings AI Journal said top performers are able to differentiate themselves by implementing managed,value-oriented AI project development pipelines.“They evaluate the potential ROI,rapidly prototype to pr

148、ove the concepts,and use disciplined data engineering processes,”she says.Disciplined data engineering spans data cleansing,ModelOps for model maintenance and retraining,and system monitoring.“That complex orchestration is required to industrialize AI at speed and at scale,”Holub says.The where,what

149、,and who of AI22|DATA+AI RADAR 2022External Document 2022 Infosys LimitedDeep learning helps achieve AI at scaleDeep learning an advanced form of AI that imitates how humans learn holds great promise for advanced AI capabilities and delivering AI at enterprise scale.Infosys Tech Compass Research fou

150、nd that companies can move out of the lowest tier of AI maturity when they integrate deep learning or related advanced algorithms into more than 30%of their AI systems.16 Data+AI Radar found evidence that this benchmark also links to good business results and drives companies toward AI at scale.Embr

151、acing deep learning draws enterprises closer to AI at enterprise scale,because the two share common technical and data requirements.First,both rely on scalable computing power.Overall,our survey respondents are evenly split on where they prefer to host third-party AI solutions,four out of five compa

152、nies engaged in deep learning prefer cloud-based AI tools to on-premises AItools.Deep learning and AI at scale both also require massive volumes of data.Some 70%of companies heavily engaged in deep learning also extensively share data,our research found(Figure 14).Most respondents in our survey are

153、on this path Two-thirds use deep learning in more than 30%of systems(Figure 15);most of that group(73%)share data extensively and most(79%)use cloud for AI(Figure 16).Companies in different industries showed more variation in data and AI practices,capability and satisfaction.On a region-to-region ba

154、sis,appetite for cloud and levels of advance AI adoption showed some distinction.AI practitioners in Australia,New Zealand,France and Germany showed a slight preference for third-party AI tools in the cloud,and a greater percentage of deep learning in their AI systems.To a large extent,deep learning

155、 and other forms of advanced algorithms are what corporate executives mean when they talk about AI today.Thats led to a gap between the reality of contemporary AI and the expectation of what companies think they can easily achieve,Agarwal says.“When you give customers an AI enabled product,they expe

156、ct it to basically do magic and suddenly solve all their problems,”he says.“The AI is there to make humans smarter and safer and better,and thats what it can do practically.”The misinformed notion that AI is practical magic is starting to change,he said,as more managers and IT staffers grow familiar

157、 with what AI can practically do.“There is more excitement from people familiar with AI.They understand that it is not a silver bullet,”he says.Figure 14.Deep learning correlates with extensive data sharingFigure 15.Two-thirds of respondents use deep learning in 30%or more of their AI systems1%73%99

158、%27%30%deep learningLimited data-sharingExtensive data-sharing30%deeplearning,67%Proportion using deep learning in more than 30%of systemsDATA+AI RADAR 2022|23 External Document 2022 Infosys LimitedWhats at stake:Data+AI must prove its nothing to fearThis is the state of AI in the corporate world:It

159、s everywhere and not often a differentiator.Most AI practitioners are not satisfied with or trusting of AI outcomes and neither is the public.17 Advanced AI requires a new way of thinking about technology and business.It requires trust:Trust in your own and others data management,and trust in AI mod

160、els.Clear data processes,strong ethics and bias management build trust.The right choices of where to apply AI and the right mix of advanced AI and AI-cloud systems can reinforce and improve AI results,for both companies and the greater good.Around the globe,governments are trying to guard against da

161、ta misuse and autonomous AI running amok.China and California have followed Europes General Data Protection Regulation with their own data-privacy rules.18 Business groups and consumer advocates in the US want to see federal data privacy legislation enacted.19 Both the US and Europe have regulatory

162、sights set on AI.European leaders have been debating the proposed EU AI Act for more than 12 months.20 On Oct.4,2022,the White House Office of Science and Technology Policy(OSTP)published a Blueprint for an AI Bill of Rights.21 The blueprint covers five big areas:safe and effective systems;algorithm

163、ic discrimination protections;data privacy;notice and explanation;and human alternatives,consideration,and fallback.Large enterprises must demonstrate that they can get value from data and AI while upholding ethical and legal standards.If they dont,they face increased compliance and reputational ris

164、ks.“Due to their significant influence on society,large enterprises have a distinct responsibility to deliver explainable,responsible AI.By explaining their data and AI outputs with transparency,stakeholders(including customers)are more likely to trust their intent,as well as their products and serv

165、ices.”Nandan NilekaniChairman,Infosys“Explaining the outcomes of the AI will continue to require human ingenuity and the inclusion of a greater diversity of perspectives in training algorithms.”Sunil SenanSenior vice president,data and analytics,Infosys24|DATA+AI RADAR 2022External Document 2022 Inf

166、osys LimitedReduced to their simplest forms,data is quantities and AI is math.Data+AI is a formula written to solve for business value.Advanced AI requires enterprises to be precise in their numbers,calculations,and equations.Data must be true,accessible,and shareable.Advanced AI requires zettabytes

167、 of data and the wisdom to understand which bits and bytes are most instructive for the present business problem.The wrong digits and right computation will not yield an accurate,trustworthy result.Our study found that AI models must be informed by bias management and ethical AI practices,and they m

168、ust bring explanation.This is critical to build trust and ensure that AI systems get used.If you cant show your work,you cant persuade skeptics to believe your solution.AI teams must clearly frame and explain both their definition of business value and choice of variables used in an AI model(or form

169、ula).This requires the perspectives of data scientists,industry specialists,and business executives.A model or formula developed without business value can only deliver a solution in search of aproblem.“You should know that an automated system is being used and understand how and why it contributes

170、to outcomes that impact you.”A Blueprint for an AI Bill of RightsThe White House Office of Science and Technology PolicyRecommendations:Precision,trust,business value Companies that think differently about data and AI will get the most business value out of AI.Heres how:1.Get your data right,and sha

171、re it.Focus on data sharing capabilities and hub-and-spoke data management.2.Build trust in advanced AI.Strengthen ethics,bias management,deep learning,AI cloud,and scaling across the organization.3.Compose an AI team biased to business value.Business leaders matter as much as data scientists.DATA+A

172、I RADAR 2022|25 External Document 2022 Infosys Limited1.Get your data right,and share itGet your data right,and share it.Focus on data-sharing capabilities and hub-and-spoke data management.Before putting any AI into production,companies must have accurate,organized data.This has long been true,but

173、the method for achieving that has evolved.The old process extract,transform,and load data into a private warehouse faced limits.Followers of that procedure can only apply AI to the data contained in the four walls of their warehouse.Data management strategies that foster data sharing,both importing

174、in and sharing out,expand the universe of available data.Complex data and AI capabilities should leverage the best features of centralized and federated data management strategies,Infosys data strategist Nayar says.“The best way to get control of your data is to centralize it.But as companies move t

175、oward AI-driven initiatives,theyre finding centralization is not the answer because the most pertinent data is all over the place,”he says.Rafee Tarafdar,Infosys chief technology officer,recommends“a hub-and-spoke strategy where companies centralize platform and technology but give teams flexibility

176、 to operate on their own.”This approach takes the best from centralized and federated and melds data management strategies.The hub is a common repository that defines the data and its location.The spokes,figuratively radiating from the hub,contain and share data as dictated by business specialists a

177、nd AI teams.Companies are increasingly establishing a hub as a common tech layer that organizes data and connecting spokes that empower business units to manage their own data.“Business units are saying,Data and AI is my asset,”Nayarsays.Sameli Menp,the chief data officer(CDO)at Finland banking and

178、insurance firm OP Financial,embraces this sort of strategy.He describes his CDO role as more of a library architect than a librarian.“We provide the library building,electricity,water,and the shelves for books.We do not take responsibility for which books are on the shelves.We do not take responsibi

179、lity for indexing the books.We provide the indexing service and 26|DATA+AI RADAR 2022External Document 2022 Infosys Limited“A hub-and-spoke strategy where companies centralize platform and technology but give teams flexibility to operate on their own.”Rafee TarafdarChief technology officer,Infosys s

180、olution,but we do not do the actual indexing,”he says.“If the books are worn out or quality is bad,its the business owner who needs to fix that.The data owner is responsible for that quality and they also need to know who they loan the books to.”Menp rejects the notion that the CDO should fix and cl

181、ean all the companys data.Rather,each individual business unit should own their own data,with the data unit providing data warehouse infrastructure and a cloud-based data-science workspace.Centralized structuring of data followed by distributed data analysis is a hallmark of companies engaged in Cen

182、tralized,56%Centralized,49%Mixed,22%Mixed,19%Data management strategyFederated,22%Federated,32%Centralized,58%Centralized,27%Mixed,22%Mixed,8%Data management strategyFederated,20%Federated,65%extensive data sharing,our research shows.More than half of companies in our highest tier of data sharing ar

183、e pursuing a centralized-to-federated data management trajectory.Technology and insurance companies survey share data the most,and they are also the two industries that showed the strongest preference for advancing from centralized to federated data management(Figure 16).Think of changes in data man

184、agement strategies as a maturity curve or a cycle,Nayar says.Data thats not organized must be centralized.Once thats established,a shift to federated data analysis brings business executives into the data+AI equation.Figure 16.Insurance and technology industries show strong preference to move to fed

185、erated than do other industriesDATA+AI RADAR 2022|27 External Document 2022 Infosys Limited2.Build trust in advanced AI“Companies have a lot of value in their data that they have not unlocked.This is leading to a realization that they cannot approach data and AI as individual use cases or projects.T

186、hey must establish an enterprise-wide strategy to discover,democratize and de-risk AI implementation at scale.”Balakrishna DRExecutive vice president and head of AI and automation,Infosys Build trust in advanced AI.Strengthen ethics,deep learning,AI cloud,and scaling across the organization.When emp

187、loyees trust that their AI systems are operated responsibly,they are more likely to work with AI outputs,experts say.This grows increasingly critical as companies put more advanced AI systems to work.AI that cannot quickly and clearly be explained ends up unused and leaves the people receiving its o

188、utputs dissatisfied.Fortunately,strong ethics and bias management practices can increase trust and satisfaction in data and AI.Specific advanced AI capabilities and practices can work together to create a virtuous cycle that leads to more advanced AI.Deep learning is an example of this.“The foundati

189、onal aspect of doing deep learning is to get tons of different kinds of data,and you can only do that in an environment where data is conveniently shared,and thats on the cloud,”Bhattacharjee says.More AI in cloud correlates with greater data sharing.Expanding deep learning and data sharing each hav

190、e a positive influence on corporate profits,our analysis shows.Sharper data honed by robust data sharing leads to better outputs from deep learning.Recall that most of the high-satisfaction data and AI use cases deliver signal detection or automation functions in essence,narrow solutions.Three-quart

191、ers of our survey respondents say they want something more broad:advanced AI at scale.Advanced algorithms such as deep learning,AI cloud,and extensive data sharing are the capabilities guide companies to better business outcomes,and a shot at scale.28|DATA+AI RADAR 2022External Document 2022 Infosys

192、 Limited3.Compose an AI team biased to business value“The mission of the AI team is to deliver democratized tools to employees,customers and partners to enable them to develop and use data responsibly,accurately and ethically across their entire ecosystem.”Andrew DuncanCEO and managing partner,Infos

193、ys Consulting Compose an AI team biased to business value.Business leaders matter as much as data scientists.Good AI teams typically involve multiple disciplines.Regardless of the question,three groups of people should always be on the AI team:data scientists,experts in the business problem,and seni

194、or executives.Its obvious that data scientists should be involved in AI work,and our study backs this up.But experts with intimate knowledge of the business problem at hand matter just as much,our analysis shows.Business experts are critical to framing out AI models and identifying the most instruct

195、ive data.Including a key subject matter expert has the dual benefit of properly training a new AI model and establishing trust in the organization when the AI enters full-scale operation,says Richard Donaldson,vice president of digital transformation at Duke Energy Corp.“Youre going to let a compute

196、r make a decision that a human once made.Youre only going to be successful if you include that decision maker on the front end of the process,”he says.Duke included these decision makers when its data team built an algorithm to prioritize work at a group of its generation stations.In the testing pha

197、se,people reviewed all the results from the algorithm,and delivered feedback when the model made a mistake and when it prioritizedcorrectly.“Youre giving that feedback into your machine learning and its actually getting smarter and smarter every day,”hesaid.Dukes AI teams focus on building trust as

198、they scale AI.“Its not enough to solve it in one business unit and then everybody else will jump on board.You have to build that trust in every single use case,”Donaldson says.Data+AI doesnt belong only to data scientists.Senior executives play critical roles in keeping AI efforts focused on core st

199、rategies.Experts in the business problem guide data scientists and AI systems in properly framing the problem at hand.A diverse,business-focused team keeps Data+AI tethered to business priorities.Senior executives are critical to scaling AI because of their knowledge of strategic operations.And when

200、 the boss is absent from the AI team,it can be costly.Companies that never involved senior executives on their AI teams reported significantly lower profit growth and revenue growth.On average,this was more than 10 percentage points lower than those who often use senior executives on their AI projec

201、ts.Once again,in their simplest form:Data is numbers.AI is math.Both must be accurate.The AI team must develop formulas to solve two questions simultaneously:“What value do you seek?”and“Who feels the impact of thatvalue?”DATA+AI RADAR 2022|29 External Document 2022 Infosys LimitedAppendix:Research

202、approach Infosys commissioned an independent third-party survey of 2,504 AI practitioners.In addition to questions about data,AI and technology practices and capabilities,we asked survey respondents for financial details including revenue range and yearly revenue and profit growth rates.The survey w

203、as conducted in May to July 2022.It included respondents from companies with more than$500 million in annual revenue in the United States,United Kingdom,Germany,France,Australia and New Zealand.IndustryFinancial services248Automotive228High tech200Healthcare221Energy,mining,or utilities191Life scien

204、ces191Manufacturing200Telecom199Insurance196Retail orhospitality213Consumerpackaged goods(CPG)215Logistics orsupply chain202We identified and analyzed a large set of actions that could affect profit and revenue change related to data and AI.We then set base cases and found via linear regression 23 a

205、ctions(of 69 analyzed)that showed evidence of a statistically significant impact on profit or revenue growth.The$467 billion in potential profit growthderives from a 10%increase in profit growth that can be achieved from 13 actions with statistically significant uplifts.Job functionInformation techn

206、ologyOperations or supply chainSales or marketingHuman resourcesResearch and developmentAccounting or finance31632132738845270072%outside of IT30|DATA+AI RADAR 2022External Document 2022 Infosys LimitedCountryAnnual revenueRole in AIEvaluationStrategyImplementation26%26%48%Role in AI20%$500 milliont

207、o$999 million20%$1 billionto$3 billion19%$3 billionto$5 billion41%$5 billionCountryUnited States30%France17%Australia or New Zealand8%Germany20%United Kingdom25%DATA+AI RADAR 2022|31 External Document 2022 Infosys LimitedReferences 1.Worldwide Spending on AI-Centric Systems Will Pass$300 Billion by

208、2026,According to IDC,Michael Shirer,Sept.12,2022,International Data Corporation.2.Why the iBuying algorithms failed Zillow,and what it says about the business worlds love affair with AI,John Cook,GeekWire,Nov.3,2021.3.Opendoor company financials,C,accessed Oct.17,2022.4.AI:Its Not Just For the Big

209、FAANG Dogs Anymore,Alex Woodie,Datanami,June 11,2021.5.KRTI 4.0 The Next Generation AI Framework for Operational Excellence,Infosys,May 24,2019.6.Expectations of artificial intelligence and the performativity of ethics:Implications for communication governance,Aphra Kerr,Marguerite Barry,and John D.

210、Kelleher,Big Data&Society,April 30,2020.7.What are the most important advances in AI?,AI100:A One-Hundred-Year Study on Artificial Intelligence,Stanford University,Sept.16,2021.8.Open Data Science APAC Virtual Conference 2020,Keynote,Shailesh Kumar,Dec.8,2020.9.Automation Journey of Bupa Global,Abhi

211、ram Mahajani,Automation Journey of Bupa Global,The Applied AI Podcast,Dec.22,2021.10.The Live Enterprise:Create a Continuously Evolving and Learning Organization,Jeff Kavanaugh and Rafee Tarafdar,Infosys Knowledge Institute,2021,p.129.11.Data Is the New Oil,Clive Humby,ANA Marketing Maestros,Novembe

212、r 2006.12.Capitalizing on the data economy,MIT Technology Review Insights,Nov.16,2021.13.Building tomorrows telecommunications network today,MIT Technology Review Insights,June 15,2022.14.Applied AI:For a diverse,more human enterprise,Mohit Joshi,Infosys Insights.15.Global Economic Outlook 2023:Glob

213、al Recession Fears Resurface,Risks Mount,The Conference Board,Oct.10,2022.16.Pervasive Intelligence for a Live Enterprise,Infosys Tech Compass,2022.17.Managing expectations of artificial intelligence,Nature,Jeff Hecht,Nov.28,2018.18.Cross-Border Data Transfers:PIPL vs.GDPR vs.CCPA,Allison Kutner,et

214、al.,Cooley LLP,April 11,2022.19.What Should and Should Not Be Included in a National Privacy Bill,Jordan Crenshaw,U.S.Chamber of Commerce,Sept.13,2022.20.The EU AI Act will have global impact,but a limited Brussels Effect,Alex Engler,Brookings Institute,June 8,2022,21.Blueprint for an AI Bill of Rig

215、hts,White House Office of Science and Technology Policy,Oct.4,2022.32|DATA+AI RADAR 2022External Document 2022 Infosys LimitedContributorsThe Knowledge Institute would like to thank the following people for providing guidance,insight and feedback throughout the Data+AI Radar research project.We are

216、grateful to be surrounded by intelligent professionals in and outside our organization that know the value of sharing data.Bill Sniffen,Cigna;Richard Donaldson,Duke Energy;Rich Cooper,Fusion Risk Management;Adriano Koshiyama,Holistic AI Inc.;Hajo von Reinhardt,Kao;Dr.Tamara StClaire,Vikas Agrawal,Or

217、acle;AustinGunter,John Bohannon,Primer AI;Saurav Agarwal,Siera.AIAisha Mendez,Akhilesh Mairpady Rao,Alec Boere,Arunkumar Ranganathan,Ashok Panda,Asit Karapurkar,Balakrishna DR,Bonnie Holub,Brent Field,Daniel Farenzena,Deepak P.N.,Dr.Ravi Kumar G.V.V.,Erik Newlin,Gary Bhattacharjee,Gaurav Bhandari,Ga

218、urav Bansal,Harald Gunia,Jagadamba Krovvidi,Jinu Koshy,Jon Warren,Joseph J.Alenchery,Kamalkumar Rathinasamy,Kamlesh Kumar,Karthik Andhiyur Nagarajan,Madhav Timalsina,Mohammed Rafee Tarafdar,Mythreya Reddy Kotha,Nandha Kumar Muthusrinivasan,Narendar P.,Prabhat Kaul,Praveen Gonabal,Pushpak Banerjee,Ra

219、jeev Nayar,Rajeshwari Ganesan,Rajnish Sharma,Ramgopal Natarajan,Randy Chambers,Ravishankar Shetty,Ritesh Saluja,Ryan Elmore,Saibal Samaddar,Sandeep Roy,Sankar S.D.V.,Satish Doreswamy,Satish H.C.,Serge Plata,Seth Siegel,Shashank Mathur,Shatam Bhattacharyya,Soumya Nandy,Srinivas Gopal Sugavanam,Sriram

220、 Sundar,Sudarshan Gopalan,Sunil Senan,Syed Ahmed,Ulku Guneysu,Uttam C.N.Ritesh,Vishwanath Taware DATA+AI RADAR 2022|33 External Document 2022 Infosys LimitedAuthorsChad Watt|Infosys Knowledge Institute,DallasJeff Kavanaugh|Infosys Knowledge Institute,Dallas Analysis and ProductionIsaac LaBauve|Infos

221、ys Knowledge Institute,DallasJitesh Gera|Infosys Knowledge Institute,BengaluruRyan Mathew|Infosys Consulting,Bengaluru34|DATA+AI RADAR 2022External Document 2022 Infosys LimitedAbout Infosys Knowledge InstituteThe Infosys Knowledge Institute helps industry leaders develop a deeper understanding of b

222、usiness and technology trends through compelling thought leadership.Our researchers and subject matter experts provide a fact base that aids decision making on critical business and technology issues.To view our research,visit Infosys Knowledge Institute at or email us at .2022 Infosys Limited,Benga

223、luru,India.All Rights Reserved.Infosys believes the information in this document is accurate as of its publication date;such information is subject to change without notice.Infosys acknowledges the proprietary rights of other companies to the trademarks,product names and such other intellectual prop

224、erty rights mentioned in this document.Except as expressly permitted,neither this documentation nor any part of it may be reproduced,stored in a retrieval system,or transmitted in any form or by any means,electronic,mechanical,printing,photocopying,recording or otherwise,without the prior permission of Infosys Limited and/or any named intellectual property rights holders under this document.Stay ConnectedI|NYSE:INFYFor more information,contact