无 Cookie 世界中的营销分析 - 大规模媒体组合模型.pdf

1、2024 Databricks Inc.All rights reservedMARKETING MARKETING ANALYTICS IN ANALYTICS IN A COOKIELESS A COOKIELESS WORLDWORLDCorey AbshireCorey AbshireJune 13,2024June 13,202412024 Databricks Inc.All rights reservedCorey AbshireSr.Specialist Solutions Architect2YOUR SPEAKERSYOUR SPEAKERS2024 Databricks

2、Inc.All rights reserved Why MMM?How do we build it?3OVERVIEWOVERVIEWUse the subtitle to add extra relevant context if neededUse the subtitle to add extra relevant context if needed2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved4WHY MMM?WHY MMM?2024 Databricks Inc.All r

3、ights reserved2024 Databricks Inc.All rights reserved5WHY MMMWHY MMMMMM platform uses pre-aggregated data at various levelsEliminating the reliance on user data and allows for calibration using external benchmarks.Online+offline marketing channelsMedia mix models help overcome key challenges in mark

4、eting analyticsMedia mix models help overcome key challenges in marketing analytics52024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved6CHALLENGESCHALLENGESUpstream Data PipelineTeam SilosMLOpsCommon implementation challenges and concernsCommon implementation challenges an

5、d concerns62024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reservedStreamline data pipeline Improving Data Quality and LineageFeature Store enforces best practices of feature engineeringFirst class Databricks ML featuresAutoML accelerate model developmentEnsure ML Feature data

6、lineageProvide access to whole media historical dataUse Databricks to accelerate the time to bring new data projects from POC to ProductionDistribute training code:single node DS code running in parallel resulting time saving and efficiency gainUnify upstream and Unify upstream and downstream data s