Data sources
Last updated 2026-10-30
Groundwork does not invent data. Every number in an analysis or a packet comes from one of the sources below, and every provider's attribution requirement is met on this page, in the application and in the exported files. If a provider is unavailable, the analysis says so instead of filling the gap.
| What | Source | Notes |
|---|---|---|
| Demographics | US Census Bureau, American Community Survey 5-year estimates, block-group level, via the Census Data API | Population, income distribution, education, labour force, occupation, industry, housing. Median household income is computed from the full income distribution across the trade area, not averaged; median age and median home value are weighted averages of the block-group medians. |
| Block-group geometry, municipal boundaries | US Census Bureau TIGERweb | Which block groups fall inside a trade area; the official city, township or CDP boundary for City mode. |
| Drive-time, walk and bike trade areas | openrouteservice (HeiGIT), built on OpenStreetMap road data | Isochrones are cached; when the provider is unavailable the analysis falls back to a straight-line radius and says so. Every report names the provider that drew its area. |
| Competitors, co-tenancy, places workbook | TomTom Search (business listings), with OpenStreetMap (Overpass) as the fallback | TomTom listings include phone and website. On this server the scan currently runs on TomTom. |
| Geocoding and address typeahead | Nominatim and Photon (komoot), both on OpenStreetMap data; the US Census Bureau geocoder for street addresses in the free check | Address to coordinates and back. |
| Basemap | OpenFreeMap vector tiles (OpenMapTiles schema) rendered with MapLibre GL | Map data from OpenStreetMap. |
Attribution
- US Census Bureau. This product uses the Census Bureau Data API but is not endorsed or certified by the Census Bureau. Terms: census.gov/data/developers/about/terms-of-service.
- OpenStreetMap. Map data © OpenStreetMap contributors, available under the Open Database License (ODbL). This covers the road network behind drive times, the fallback business listings, geocoding and the basemap.
- TomTom. © TomTom. All rights reserved. Business listings are proprietary to TomTom and its suppliers, licensed for use within Groundwork; see TomTom's copyright notice. Where TomTom results include OpenStreetMap-derived data, the OpenStreetMap credit above applies.
- openrouteservice. © openrouteservice.org by HeiGIT | Map data © OpenStreetMap contributors.
- OpenFreeMap. © OpenFreeMap © OpenMapTiles, data from OpenStreetMap.
- Nominatim and Photon. Search © OpenStreetMap contributors; Photon by komoot.
How we test the verdict
The verdict is checked against real site decisions, and we publish what we find, including where it falls short.
- The test. On one brand's 73 decided markets, scored on demographic signals only (that history has no competitor or anchor counts), the verdict cut-offs used live (Approved at 68, Conditional at 58) agree with the team's past decisions 77% of the time when each market is held out in turn. Always answering "Approved" would agree 71% of the time.
- The honest caveat. On five random splits that hold back 30% of the markets, the verdict agreed 75.5% of the time on the held-back markets, against 77.3% for always answering "Approved" on the same markets. At this sample size the difference is within chance, so read the verdict as consistency with past decisions, not as a measured accuracy, and never as a forecast of store sales.
- How it gets better. Each decided site can carry its outcome (opened above, on or below plan, closed). That is how the verdict gets tested against results rather than against past calls. Settings runs the same tests on your own workspace once it has eight decided markets, and shows the numbers to you.
What we do with it
- The data is pulled per site when you run an analysis and cached in your workspace so re-opening a site is instant. Cached results are yours to delete.
- Exports (HTML and Word reports, the comparison workbook, the overview map, the places workbook and the packet's README) carry the attribution lines above so the files can travel on their own. The comparison CSV is plain data for spreadsheets and carries none; inside a packet it travels with the README.
- Groundwork does not redistribute the underlying datasets in bulk and asks customers not to extract them from exports for that purpose.
- US only: the American Community Survey covers the United States and Puerto Rico; Groundwork does not score sites elsewhere.
Questions
Spotted a wrong number or an attribution we missed? Email christian@groundworkintelligence.net.