Everything in its Right Place: Optimising Ranking in Online Grocery
summary
Mobile grocery ranking combines three concrete pieces: a next-basket repurchase probability model (CARP) trained on delivery histories and product periodicity, a position-bias corrected STAR score that estimates true search-term–article relevance, and an explainable ranking templates system that linearly mixes CARP STAR and business boost signals. Implemented for an 8,000‑item catalog with 30 shown per purchase page, it runs in batch today, A/B tests show higher conversion and lifts in mAP@30 MRR and NCG, and real‑time updates remain work in progress.
AI-written summary · about
chapters (8)
- 0:00Picnic and ranking intro
- 2:14Ranking in Picnic app
- 5:23Ranking templates per section
- 7:09Repurchasing and CARP model
- 15:57Search relevance and bias
- 21:16Balancing relevance and boosts
- 24:48Ranking strategies and future work
- 25:50Q&A
related talks
37mFood is personal! Design and impact of personalized recommendations in e-grocery
44mRobust AI Search Ranking for Radical C2C Marketplace Growth
38mThe Power of the Bonus Card: Road to Personalised Search
41mHow Deliveroo improved the ranking of restaurants
35mLearning to Judge
55mImproving product discovery via relevance and ranking optimization
21mAmazon Search: The Joy of Ranking Products
25mBetter search relevance using Learning to Rank at mobile.de