Turn site selection into a testable market thesis

How a site screen is built

1. Define the operating modelDescribe the concept, target customer, format, minimum demand, labor needs, distribution radius, rent tolerance, and decision stage.
2. Choose the geography ladderScreen broad markets first, then counties, cities, tracts, or drive-time areas. Do not treat administrative boundaries as actual customer trade areas without justification.
3. Measure demand and supplyCompare population, household income, category-relevant households, employment, establishment counts, and change over time. Normalize counts so large places do not win by size alone.
4. Add operating constraintsInclude wages, labor availability, housing pressure, logistics, and site-specific costs where reliable data exists.
5. Test the modelInspect outliers, alternative weights, missing variables, boundary changes, and small denominators. Show which inputs drive each candidate.
6. Move from screen to diligenceValidate finalists with current leases, parcel data, zoning, traffic, competitors, customer research, and site visits.

Worked-example framing: Austin restaurant entry

The Austin sample begins with a five-county metro screen. It compares county population and income, metro employment and wages, restaurant establishment and job growth, mixed-beverage receipts, price context, and a tract-level composite.

The initial evidence points toward the Williamson-Hays suburban ring for further investigation. Between 2021 and 2025, the QCEW values retained in the brief show Williamson and Hays together adding 276 restaurant and bar establishments, compared with 228 in Travis. The cited mixed-beverage series shows Austin trailing-12-month receipts down 0.6% through May 2026 even as the number of reporting establishments increased. Those signals suggest a suburban-growth and urban-core-saturation hypothesis.

But the tract score also demonstrates why a shortlist needs challenge. Its top-ranked tract’s 1,241% population growth came from an estimated increase of only 17 to 228 residents. Another leader ranked highly because of extreme density despite low median household income. Re-running the model with absolute growth, outlier treatment, different weights, and real operating constraints is part of the work, not an optional afterthought.

Representative primary sources

• Demographics and housingU.S. Census Bureau ACS tables for population, households, income, commuting, housing, and workforce characteristics.
• Industry activityBLS Quarterly Census of Employment and Wages for establishments, employment, and wages by industry and geography.
• Business countsCensus County Business Patterns for establishments and employment by industry.
• Geographic boundariesCensus TIGER/Line shapefiles for compatible geographic identifiers and shapes.
• Labor conditionsBLS Local Area Unemployment Statistics for local labor force, employment, and unemployment.
• Regional costsBEA Regional Price Parities for metro and state price-level comparisons.
• Road trafficFHWA Highway Performance Monitoring System for annual average daily vehicle traffic on covered road segments. Traffic counts are not pedestrian foot traffic, store visits, or proof of customer demand.
• Hazard and environmental screensFEMA National Risk Index and EPA Facility Registry Service listings for early-stage risk review. Modeled area risk and program listings do not replace parcel flood, title, engineering, or environmental due diligence.
• Weather historyNOAA GHCN-Daily for station observations and historical normals. These are not site-specific forecasts or guarantees about extreme events.
• Federal lease contextGSA federal lease inventory for public-sector occupancy and rent context. Federal, often fully serviced negotiated leases are not private asking rents or exact property comparables.

What public data cannot settle

A tract’s residents are not necessarily a location’s customers. Business listings may be stale or classified inconsistently. Official releases can lag current construction, leases, closures, traffic patterns, and competitor moves. A model also cannot infer concept fit, management quality, unit economics, or contractual risk. Use the screen to focus diligence, not to skip it.