Predicting Store Cannibalisation using ML
summary
Model predicts percentage spend per LSOA for a hypothetical new store using LightGBM regression trained with leave‑one‑out on 19 recent openings, ingesting loyalty card postcodes plus drive times, IMD, population density, competitors, store size and retail centre category. It builds catchments by rolling up predicted spend shifts and runs fast — about 100 postcodes in 15 minutes — with ~1.5% error and outperforms a gravity model in 80% of cases.
AI-written summary · about
chapters (11)
- 0:00Introduction to project
- 0:51Store cannibalisation challenges
- 2:16Why not gravity models
- 3:21Modeling approach details
- 5:52Training and validation
- 6:36Results for Bangor
- 7:40Results for Saffron Walden
- 8:20Results for Scotland
- 9:02Model performance summary
- 9:22Case study Epsom Bridge
- 9:44Conclusion and benefits
related talks
40mMachine Learning over Small Data: generating new variables with predictive power
30mAI in Retail - Advancing Promo Demand Forecasting
28mMastering Demand Forecasting: Lessons from Europe's Largest Retailer
24mUnlocking revenue: newcomer classification for boosted profits
22mFrontiers of Machine Learning from the Lens of Retail
08mMachine Learning for Bangladeshi Retailers
36mCustomer Lifetime Value: Models, Metrics and a Multitude of Uses
31mFrom stockouts to happy customers: Proven solutions for time series forecasting in retail