当前位置:首页 > 报告详情

GenAI 在客户服务中的可观察性.pdf

上传人: Fl****zo 编号:718692 2025-06-22 32页 1.84MB

1、GenAI Observability in Customer Care EarnInWillem DHaeseleerMatteo CiccozziAgenda3Customer Support At EarnIn5 minutesChatbot Journey5 minutesMetrics5 minutesPlatform10 minutesLooking Forward2 minutesHallucinations8 minutesEarnIn4Make Any Day PayDayMake Any Day PayDayNo interest,no credit checkOver 1

2、9 million downloads and countingCustomer Care5At EarnInTaxonomy of 100+customer support issues24/7 real time support across USHighly regulated by CFPBCustomer Care6At EarnInTaxonomy of 100+customer support issues24/7 real time support across USHighly regulated by CFPB7March 2024FAQ RAGJune 2024Tool

3、callingAugust 2024Datadog-DatabricksOctober 2024Ai/Bi Dashboard February 2025 Agentic CapabilitiesApril 2025Multimodal2025+ReasoningChatbot JourneyAt EarnInMetricsImmediate feedback during experimentationMetricsHuman reviewEvolving definition of metricsHigh cardinality metrics8Closing the feedback l

4、oopMetricsSingle use case evaluationDid we help the customer?Did we contain the conversation?Intent precision Multi use case evaluationMeasured over all chat trafficBottom Line/North starIntent recallContainmentCoverage9Measuring in productionMetrics10Containment/CoveragePlatform12PlatformArchitectu

5、re13PlatformArchitecture14PlatformArchitecture15PlatformArchitecture16PlatformArchitecturePlatform17Publishing Closeup*https:/microservices.io/patterns/data/transactional-outbox.html*Platform18Materialization CloseupPlatform19Materialization Self-ServePlatform20Materialization Self-ServePlatformClou

6、deventsTransactional OutboxDefine contract for eventsPublish events to kafkaDelta Live TablesAccess control w/UnityNear real time optionSelf-Serve Materialization21Developer ExperienceProtobuf or JSON SchemaSchema Registry&Lifecycle Integrated data environmentNotebook explorationTrack experimentsInc

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文主要介绍了EarnIn公司在客户关怀领域采用GenAIObservability的实践和规划。关键点如下: 1. EarnIn服务:提供“Make Any Day PayDay”服务,无利息、无信用检查,下载量超1900万。 2. 客户关怀:拥有100多种客户支持问题分类,提供美国范围内24/7实时支持,受CFPB严格监管。 3. 评估指标:实验期间提供即时反馈,关注单一和多用途案例的评价,如客户帮助、对话控制等。 4. 平台架构:使用Cloudevents、Delta Live Tables等技术,实现事件合同定义、数据环境集成等。 5. Ai/Bi仪表板:用于跟踪实验、事件响应等,强调多模态和推理能力。 6. 不足与展望:指出需要高度专业知识,理解用户行为至关重要。提出主动检测幻觉、用人类代理数据优化聊天机器人等发展方向。 核心数据:下载量超1900万,100+客户支持问题分类,24/7实时支持,CFPB监管。
"支持EarnIn的秘诀是什么?" "如何衡量聊天机器人的成功?" "GenAI观察性如何助力客户关怀?"
客服
商务合作
小程序
服务号
折叠