The data sources behind a defensible site selection model

Site selection rarely fails for lack of data. It fails because sources are joined to inconsistent geographies, refreshed at different cadences, and normalised in ways nobody documented. Here is the short list worth building on.

Demographics

In the United States, the Census Bureau's American Community Survey provides population, income, age and household composition at tract and block-group level, with five-year estimates for small geographies. In Canada, Statistics Canada census profiles and the census subdivision hierarchy serve the same role. Both are authoritative, free, and lag reality by design — treat them as structural, not real-time.

Competitors and amenities

OpenStreetMap, queried through Overpass, gives near-current points of interest by category and is unmatched for coverage breadth. Its weakness is completeness variance between regions, so use it for relative density rather than absolute counts, and sanity-check dense categories against a second source.

Accessibility

Mapping providers supply travel-time isochrones and road network routing. The important discipline is fixing one travel mode, one time-of-day profile and one duration across your whole candidate set.

Context layers

• Climate and seasonality from Open-Meteo, which materially affects footfall-driven formats.
• Labour market indicators for staffing feasibility.
• Cost of living to calibrate spend expectations between markets.

Where the joins break

Three recurring failures: mixing census vintages inside one comparison; attaching tract-level demographics to an isochrone without area-weighting the partial tracts; and counting competitors in a radius while measuring demand in an isochrone. Each produces a plausible number that cannot be reproduced.

Documenting vintage

Put the source and vintage of every input on the report. A score that names its sources can be challenged and improved; one that does not is an opinion with decimal places. Georithm publishes its full source list and shows each factor's contribution to the final score.