Bayesian New Store Site Selection and Cannibalization Modeling
Wegmans Food Markets
The Challenge
Every time Wegmans considers opening a new location, two questions are tangled together. How much will the new store sell? And how much will the nearby Wegmans stores lose? These aren't independent — a great new location might look less attractive once you account for the sales it pulls from your own existing stores a few miles away.
Most retail site selection approaches either ignore cannibalization entirely or treat it as a separate afterthought. Wegmans wanted a single model that answers both questions together, with honest uncertainty bounds, using demographic data, geographic information, and whatever could be learned from their limited history of past store openings.
Our Approach
We built a spatial model that treats each store as part of a geographic network rather than an isolated data point. Demographic and trade-area data — population density, income levels, competitive landscape — feed into sales predictions for any proposed location. The cannibalization piece works through overlapping trade areas: when a new store's catchment overlaps with an existing store, the model estimates how much traffic shifts rather than grows. The whole thing shares information across locations and demographic segments so that even with a relatively small number of historical openings to learn from, the estimates stay grounded.
Results
The model landed at 13-14% prediction error for new store sales — accurate enough to meaningfully inform site selection decisions.
The cannibalization effect came in at roughly 1%, meaning a new Wegmans typically costs nearby sister stores about a penny on the dollar. That's a small but real number, and having it quantified with uncertainty gives the real estate team something concrete to factor into their analysis. The work led directly to an ongoing advisory relationship.
PyMC Labs Team
- Luca Fiaschi
- Sef M
Let's Chat, We Respond Fast
Tell us about your problem. We typically respond within 24 hours.
Schedule a Consultation