Skip to main content

Methodology

What we measure, and what a score will not tell you

GISmetric scores U.S. markets from published federal data. This page sets out what goes into a score, which dataset and which release each input comes from, where the coverage runs out, and the questions the figures are not able to answer. The release numbers below are read from what is actually loaded rather than from what was intended.

What the score measures

An opportunity score combines 6 factors, each built from measures that express a position among comparable areas rather than a raw quantity. It runs 0 to 100 and it is a comparison, not a forecast: it says where an area sits relative to its peers on things that are known to matter, and it does not predict revenue, footfall or survival.

Positions rather than absolutes, because an absolute cannot be judged without a frame. Three thousand residents per square mile is dense for Montana and nearly empty for Brooklyn, and a product that scored both on the raw figure would be telling a Montanan their best option is poor. The comparison set is stated on every analysis.

Nationaldefault
Percentile rank against all geographies of the same level in the United States. Answers: is this area good in absolute terms?
Within metro
Percentile rank against geographies of the same level inside the same CBSA. Answers: is this area good relative to the alternatives nearby?
Geographies scored
Block group, Metro / micro area, County, Census tract, ZIP code area. Nation and state are excluded because there is nothing meaningful to rank them against: fifty-one peers at state level, one at nation.
Business types
18, each with its own starting weights and its own trade-area radius.

What is not published here

The weighting calibrated for each business type, the ranking and distance-decay functions and the arithmetic that combines the factors are proprietary. What is published is what is measured, which dataset and release it came from, where the coverage stops and what the result cannot support — the things a reader needs in order to judge whether a score answers their question. Every analysis additionally records the weighting it applied, so a customer can always reconcile a figure with the run that produced it.

The 6 factors

Each factor is scored on the same 0 to 100 scale from the measures listed under it, and each of those measures names the public dataset it is read from.

Customer Demographics

not available everywhere

How well the resident population matches the target customer profile: share of population inside the target age range, plus educational attainment and household composition.

Measures

Adults with a bachelor's degree or more, Households of the type this sector serves, Residents in the target age range.

Sources: American Community Survey, 5-year estimates (acs.B01001); American Community Survey, 5-year estimates (acs.B15003); American Community Survey, 5-year estimates (acs.B11001)

Income Match

Share of households falling inside the preferred household-income band, computed from the full income bracket distribution rather than the median alone.

Measures

Households in the target income band, Median household income.

Sources: American Community Survey, 5-year estimates (acs.B19001); American Community Survey, 5-year estimates (acs.B19013)

Population Density

not available everywhere

Residential population per square mile of land area, blended with a daytime-population proxy derived from the number of jobs located in the area.

Measures

Daytime population per square mile, Residents per square mile.

Limitation. The daytime term counts jobs, not people present. It captures a business district filling up on a weekday morning, and misses a shopping centre whose visitors outnumber its staff many times over. It is a proxy for weekday commercial activity, not a footfall measurement.

Sources: American Community Survey, 5-year estimates (acs.B01003); Cartographic Boundary Files (tiger.ALAND); LEHD Origin-Destination Employment Statistics (LODES) (lodes.WAC); LEHD Origin-Destination Employment Statistics (LODES) (lodes.RAC)

Competition

not available everywhere

Inverse of competitive pressure. Combines distance-decay weighted competitor density inside the trade area with a market saturation measure comparing observed establishments against the number expected for the area's population.

Reads backwards. Its measures rank how crowded an area is, and the factor inverts them — so a high score here means a market that is less contested, not more.

Measures

Competitors, weighted by how close they are, Establishments of this kind per 10,000 residents.

Sources: Overture Maps Foundation data (overture.places); County Business Patterns / ZIP Code Business Patterns (cbp.establishments); American Community Survey, 5-year estimates (acs.B01003)

Market Growth

not available everywhere

Trajectory of the local market: population and housing-unit change between the two most recent non-overlapping ACS 5-year vintages, plus year-over-year change in the number of establishments in the sector.

Measures

Change in establishments of this kind, Housing unit change since the prior survey, Population change since the prior survey.

Sources: American Community Survey, 5-year estimates (acs.B01003); American Community Survey, 5-year estimates (acs.B25002); County Business Patterns / ZIP Code Business Patterns (cbp.establishments)

Accessibility

not available everywhere

How easily customers can reach the area. Built from road network density, proximity to arterial and highway infrastructure, road network node density, transit stop proximity, and the local commuting profile. Drive-time isochrones are deliberately not used in v1.

Measures

Centreline road miles per square mile, Commuters travelling 45 minutes or more, Distance to the nearest arterial road, Distance to the nearest transit stop, Households with no vehicle, Road network nodes per square mile.

Limitation. No drive-time isochrones in v1. A national routing engine (Valhalla/OSRM) needs roughly 50 GB of disk and more RAM than the 8 GB production host can spare alongside PostgreSQL and the application. Isochrones are deferred to a later phase, either scoped to a single region or delegated to an external routing service.

Sources: Cartographic Boundary Files (tiger.roads); Overture Maps Foundation data (overture.places); American Community Survey, 5-year estimates (acs.B08303); American Community Survey, 5-year estimates (acs.B08201)

How the factors are weighted

The factors do not count equally, and how much each one counts depends on the business. Foot traffic decides a coffee shop and household income decides a furniture showroom, so every one of the 18 business types starts from its own weighting rather than from a single house view of what makes a good location.

That starting weighting is a recommendation, not a verdict. It can be changed on any analysis, the ranking updates as it moves, and the weighting that produced a result is stored with the result — so a figure in a report can always be traced back to the weighting behind it, and a later change to our recommendation does not silently rewrite what an earlier analysis said.

A factor set to zero is removed rather than scored zero, which is the same distinction the next section makes about data that is missing: not counting something is not the same as counting it as the worst possible.

Data quality and missing measurements

A factor that cannot be computed for an area is dropped and the remaining weights are renormalised. It is never filled with a zero, and that distinction is the difference between two opposite claims: a zero says we measured this and it is the worst possible, a drop says we could not measure this. Every score records how many components went into it, so a five-factor score is visible as one.

It has a consequence worth stating plainly, because it looks like a bug when you meet it. Two areas can be scored from different numbers of components, and the comparison between them is then not quite like for like. Where the product knows this it says so — the side-by-side comparison refuses areas whose scores do not mean the same thing rather than footnoting them.

Survey margins of error

The American Community Survey is a survey, so every estimate it publishes carries a margin of error, and at tract and block-group level those margins are often wide. They are carried through the scoring rather than discarded, and an estimate too imprecise to rely on is labelled as such on the analysis and in the report. A number with a wide margin is still the best available answer; presenting it as though it were precise would be the problem.

Where the data is thinner

Two of the inputs do not cover the whole country, and both limits come from what this installation can hold rather than from any judgement about which places matter.

  • Named competitors and transit stations are loaded for VA, MD, DC, WV, NC. Outside those states competition falls back to establishment counts, and distance to a transit station is not measured at all — recorded as unmeasured rather than as a long distance, because those are different claims and the second would rank a well-served neighbourhood as badly connected.
  • Accessibility is computed at county, census tract, block group. It is not a property of a metro area: at that scale a road-density figure averages a downtown with a hundred miles of farmland.

This produces an asymmetry the product does not hide. A tract in Manhattan is scored on five of accessibility’s six components because transit data does not reach it, while a tract in Arlington is scored on all six — and Arlington comes out higher. That is a real limitation of the coverage, not a finding about the two places.

The national screen is a different number

Expansion screening ranks metropolitan areas, and at that scale two of the six factors cannot be computed the same way: accessibility is a property of a site rather than a metro, and competition is measured by saturation alone. The result is a five-factor score that is not comparable with what a tract-level analysis produces, and it is labelled a screen rather than an analysis for exactly that reason.

It reports two numbers per metro and never blends them. How promising a market looks and how much it resembles the one you already operate in are different questions: a metro much like home that is already saturated is a different proposition from a wide-open one whose customers are nothing like the ones your concept was built for. Averaging them would produce one ranking and destroy the only interesting information in the pair.

Sources and licensing

Release numbers are read from the database rather than from this page. If an ingest half completed, this table says so.

Every source, with the release loaded, its publisher, its licence and its publication schedule
SourceRelease
Census cartographic boundariesU.S. Census Bureau · Public domain (U.S. Government work)2024loaded August 22, 2026
American Community Survey, 5-year estimatesU.S. Census Bureau · Public domain (U.S. Government work)2024loaded August 22, 2026
County Business PatternsU.S. Census Bureau · Public domain (U.S. Government work)2023loaded August 22, 2026
LEHD Origin-Destination Employment Statistics, workplace areaU.S. Census Bureau · Public domain (U.S. Government work)2023loaded August 22, 2026
TIGER/Line national roadsU.S. Census Bureau · Public domain (U.S. Government work)2024loaded August 22, 2026
Overture Maps placesOverture Maps Foundation · Apache-2.0, CC0-1.0, CDLA-Permissive-2.0 (attribution: https://overturemaps.org/)2026-08-19.0loaded August 22, 2026

Business establishment counts are always the oldest figure in any analysis, and that is the publisher’s schedule rather than neglect: County Business Patterns appears about two years after its reference year. Whether a newer release exists is checked daily and shown on the home page.

Licences

One source is deliberately absent. Overture’s transportation theme would have supplied the road network, and a sample of it resolves entirely to OpenStreetMap and TomTom under ODbL — a share-alike licence, which would make this platform’s database a derivative work and oblige us to publish it. Road geometry comes from Census TIGER/Line instead, and a test now fails if a share-alike source is ever named as a factor input.

What this cannot tell you

All 14 of them. The 4 marked below appear on every report; the rest appear when they apply, and they are listed here because the ones that apply conditionally are the ones a reader most wants to know exist.

A score is a position, not a verdictalways shown
Every score in this report is a comparison against other areas of the same kind, not a measurement of whether a business will succeed. An area at the top of the ranking is the best of what was measured here; it is not a guarantee, and the ranking says nothing about rent, staffing, licensing, footfall at a specific address, or the operator.
Factor scores are blends, component ranks are percentilesalways shown
A component's rank is a genuine percentile among the comparable areas named beside it. A factor score is a weighted mean of those percentiles, so a factor scoring 88 is not the same claim as sitting at the 88th percentile. Both are shown so the difference is visible.
The unit is an area, not a sitealways shown
Scores describe a whole census area. Two sites a block apart inside the same area receive the same score, and the difference between them can decide the business. Use this to choose where to look, then look.
Ranked within one metro area, not nationally
These scores are positions among comparable areas in the same metro area. A score of 80 here means the top of this metro, which is a different claim from the top of the country, and it cannot be compared with a score of 80 from another metro.
Not every factor could be computed
Where a factor had no data for an area it was removed from that area's score and the remaining weights were spread over what was left, rather than filled with a zero. That keeps the score honest about what it measured, and it means a score built from fewer factors is not directly comparable with one built from all six.
Some survey estimates are imprecise
American Community Survey figures are estimates with published margins of error, and margins widen as areas get smaller. At least one input behind this report exceeds the reliability threshold, which is flagged where it occurs. Treat those figures as indicative of direction rather than as measurements.
Competitor counts cover part of the country
Competitor locations come from an openly licensed extract that covers Delaware, the District of Columbia, Maryland, North Carolina, Virginia and West Virginia. Outside those states no competitors were counted, which is not the same as none existing, and the competition factor is left uncomputed rather than scored zero.
A zero that means zero
Where saturation is reported as zero for this area, the establishment source does cover it and lists none of this kind of business. That is a finding rather than missing data, and it can mean an unserved market or one that has already been tried.
Business counts trail the population figures
County Business Patterns is published about a year behind the American Community Survey, so the establishment counts in this report describe an earlier year than the demographics. Both vintages are named in the sources section.
Network node density ranks, it does not count junctions
The road network measure counts points where separate road segments meet in the source, which overcounts real junctions because a street is split wherever its name or county changes. The ranking between areas holds; the absolute figure should not be read as a junction count.
Daytime population counts jobs, not people present
The daytime term is built from where people work. It captures a business district filling up on a weekday morning and misses a shopping centre or a campus whose visitors are not employed there.
Distances are straight lines, not drive times
Accessibility and competitor distances are measured over the ground rather than along a route. A river, a rail line or a limited-access highway between two points makes the real journey longer than the figure shown.
This ranking reflects the weighting you chose
The weights applied here are not the sector defaults. They change how the six factor scores combine into one number, and therefore the order of the ranking; they do not change any factor score. The weighting used is listed in the method section so the result can be reproduced.
How this text was writtenalways shown
The findings in this report are generated from the stored analysis by a fixed procedure, not written by hand and not written by a language model. Every sentence is composed from figures shown elsewhere in the document, which is why the wording is plain and repeats itself between reports.

Traceability and change control

Every analysis stores the weighting it applied and the release of every source that answered it. Each score can be opened down to its components, and each component shows its raw measurement in its own units alongside its position among comparable areas — so a figure can be reconciled against the run that produced it rather than taken on trust.

Reports quote the release of each source they used, and the document is stored whole. A data refresh changes what a new analysis says; it does not change what a report already said. The same applies to the weighting: a revision to what we recommend for a business type affects analyses run afterwards and leaves earlier ones as they were issued.

  • Boundaries and road network: U.S. Census Bureau TIGER/Line.
  • Demographics: U.S. Census Bureau American Community Survey 5-year estimates.
  • Business establishments: U.S. Census Bureau County Business Patterns.
  • Workplace population: U.S. Census Bureau LEHD Origin-Destination Employment Statistics.
  • Competitor locations: Overture Maps Foundation Places, CDLA-Permissive-2.0.