What this is

Every chain has a pattern it repeats. Walmart wants a highway and a big parcel. Whole Foods wants density and income. Wawa wants a commuter route. Firms like Placer.ai and Buxton sell this analysis on proprietary footfall data. This is the version you can build from public data alone — and check, because the method is published and the sources are free.

How it works

For every one of the 173,178 mapped US locations of 177 retail chains, PlazaLens measures the trade area around it: how many people live within 1, 3 and 5 miles, what they earn, how old they are, how far the nearest arterial and traffic signal sit, and what mix of other retail surrounds it. That is 1,913,197 points of interest and 239,781 Census block groups, reduced to 112 numbers per site.

Aggregate those by chain and you get a fingerprint — what a typical location of that brand looks like. Compare an address against all of them and you get a ranking: which chains that trade area most resembles, and why.

What makes it defensible

1. It is validated on markets the model has never seen. Retail is spatially autocorrelated — hold out a random Starbucks and its neighbour three blocks away is still in training, so the score flatters itself. Whole metro areas are held out instead.
2. A chain never counts itself. Every measurement around a site excludes that site's own brand. Without it, a Starbucks trade area trivially looks like the place Starbucks goes.
3. Too good is treated as a bug. The build fails if a model scores above 0.97 AUC, because at that level something has leaked. That rule caught five separate defects that would otherwise have shipped looking excellent and being worthless.
4. It beats the dumb baselines, or it says so. Every chain is compared against ranking by population alone and by a category-gap rule. 86 chains are modelled — a chain only earns a model once it has enough locations for the measurement to mean anything, and 91 of the 177 in the corpus do not clear that bar. They still count as context around everyone else's sites.

What it is not

It finds resemblance, not verdicts. There is no revenue data and no record of stores that closed, anywhere in public data, so nothing here can say a location would succeed. A chain's absence may mean untapped demand — or a site evaluated and rejected for a ground lease or franchisee territory this data cannot see. Every report says so.

The data

OpenStreetMap for locations and points of interest, US Census ACS 5-year estimates for demographics, TIGER for boundaries. All free, all public, all reproducible. No employer data and nothing under NDA. Full methodology →