Next Best Action: Maximizing Customer Value
summary
Claims a pipeline of lookalike/response/uplift classification models plus hybrid recommendation and ranking reduces manual campaign work and speeds time to market from months to weeks, with a feature store that cuts duplicated feature work and enforces consistent definitions. Describes using AB testing, monitoring, multiarmed bandits and reinforcement learning for ad feeds, and reports concrete numbers: acquisition is 5–25× costlier than retention, personalization yields 10× conversion versus basic emails, and a 50‑model next best action rollout can take 3–6 months versus years.
AI-written summary · about
chapters (16)
- 0:00Next Best Action intro
- 1:09Business growth basics
- 2:34Marketing revenue drivers
- 5:27Marketing offers beyond discounts
- 6:50Customer-centric marketing
- 8:57Bridging marketing with machine learning
- 11:45Next best action framework
- 13:31Channel optimization
- 17:01ML classification models
- 19:53ML pipeline for models
- 22:28Feature engineering sources
- 23:53Building a feature store
- 27:47Optimizing ML pipelines
- 30:39Targeting and A/B testing
- 32:27Key takeaways
- 33:11Q&A
related talks
27mEnhancing Personalization Using Recommendation Systems & Multi-Armed Bandits
20mUsers acquired - Now what?: Maximising revenue from existing players
38mRecommendations at Rappi: from MVP to Data Product
27mTransform online shopping experiences with personalized recommendations
20mWinning the Game: Leveraging Contextual Bandits for Creative Optimization
27mPreventing churn like a bandit
21mDemystifying Machine Learning for Programmatic UA & Retargeting
29mThe Science of Selling: How Data Shapes Unique Customer Journeys