Spark Ignited：使用 ML 和 Databricks 構建現代營銷團隊.pdf

《Spark Ignited：使用 ML 和 Databricks 構建現代營銷團隊.pdf》由會員分享，可在線閱讀，更多相關《Spark Ignited：使用 ML 和 Databricks 構建現代營銷團隊.pdf（8頁珍藏版）》請在三個皮匠報告上搜索。

1、2024 Databricks Inc.All rights reservedSpark Ignited:Building Modern Marketing Team Using ML and Databricks1Mateusz Mateusz UjmaUjma,Ph.D.,Ph.D.Rogers Communications,Toronto,Canada2MarTech today is shifting its focus from execution of campaigns to experience and journey of the Customer,Example of Ma

2、rTech powered Campaign Orchestration EngineCustomer Data AI models predicting Customer PropensityPersonalised TargetingOffer/Ad on FacebookThird party websiteSearches about the product/offer on GoogleRPersonalised Dynamic Ad Behind Sign-inEmail OfferPersonalised Landing PageCustomer sees offer and e

3、ngages in ChatSigns up for the OfferCustomer uses the TFN in the email and reaches Call CentreUpsell AppPurchase is complete,engage customer with relevant contentConversionData on customer interaction and result sent back to strengthen AI furtherData Driven,Customer centric approach that adjusts to

4、individual customer needs in real time and across channelsCAPABILITIES-Automated Execution&Testing Real Time Data&DetectionCustomer&Market Data AI Driven DecisionsPersonalizationScale journey orchestration for triggered comms in all channelsIntegrate event steaming for personal offer in RT on digita

5、lPresent the right offer based on customer 360Arbitrate based on MoT,AI models&Cx behaviour in Real timeCapture time sensitive Moments of Truth(MoT)data in low latency Provide online and offline moments for RT action&presentmentLeverage Customer behavioral data across RCI(RSM,TSC,etc)Integrate 3rd p

6、arty consumer data for insights&MoTsDynamic content based on Cx MoT&preferencesTailor channel comm treatment based on preferred channel Structured Data Data Science3Marketing Data Science PlatformWhyWhereWhenWhoWhatChurnModelsOffer OptimizationConverged Churn ModelsProduct Specific ModelsCustomer Se

7、gment ModelsCarrier PredictionInterbrand ModelsPrepaid to Postpaid0/0/22/0/3Black Friday/Cyber Monday Churn3/6/12 Months Churn PredictionEnd of PromotionChurn TriggersDevice Discount PredictionInsight:Retention requires over 20+models to provide full product supportRequirement:Data,model operations

8、and campaign activation need to be standardizedRequirement:Bring data together from across the business0123+Churn Rate by#of ProductsChurn RateInsight:Higher Churn for customers who did not activate Insight:Customers with multiple products are less likely to churnNot ActivatedUnknownOtherGoogle,Sams

9、ungChurn Rate by Device Manufacturer IMEIRequirement:Collate billions of network data points dailyRequirement:Rescore customer base in real-timeChurn RateInsight:Customers that visit retail churn more oftenInsight:High risk calls are time sensitive churn triggerRequirement:Use NLP/GenAI to analyze a

10、gent notes1357911 13 15 17 19 21 23 25 27 29Churn Rate by day since High-Risk CallField VisitWebsite VisitCall Center CallRetail VisitChurn Rate by InteractionRequirement:Monitor model performanceModel PerformanceInsight:Issues in input data cause intermittent model performance issues Insight:All mo

11、dels decay over timeRequirement:Monitor model inputsModel Capture Rate by Month1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Model Capture Rate by WeekMarketing Data Science PlatformData SourcesIngestion&ProcessingData ArchitectureData Science&MonitoringActivationDatabases(Customer Data)Streams(Wireless Tower Data)Files(Demographics)Web services(OpenAI)Near Real TimeBatchUnity CatalogModel Feature StoreTarget Data StoreModel Output StoreDatabricksNotebooksData DriftModel Validation