Evaluating a location data provider comes down to seven things you can test yourself: coverage in your actual markets, positional accuracy, freshness, attribute fill rates, the panel behind any movement data, provenance and consent, and license terms. Vendor-published record counts settle none of them. Your own store list does.
That gap matters because the buying process usually runs backwards. A vendor opens with a national record count, a coverage map, and a demo over a market where their data happens to be excellent. Six months later an analyst notices that a competitor the file swore was open closed last spring, and the trade area work built on top of it has to be redone.
What location data providers actually sell
Three different products get sold under the same label, and they fail in different ways.
POI records. A structured file of places, each with a name, category, coordinates, address, and a set of attributes. This is the map of what exists. Foursquare, Precisely, Data Axle, and SafeGraph all sell in this category, in flat files or through an API. If you are new to the category, our guide to POI data covers what sits inside a record.
Movement panels. An estimate of visits, built from mobile devices that share location through opted-in apps, matched to POI records and extrapolated to the full population. Placer.ai states that its panel represents roughly 9 percent of the US population, split 55 percent iOS and 45 percent Android (Placer.ai, checked August 2026). Every number a movement panel produces is a model output, not a count. We compared the major vendors in this category separately in our foot traffic provider comparison.
Analysis environments. Esri ArcGIS Business Analyst does not originate its own POI panel. It lets you select the underlying source, currently Foursquare, Data Axle, or SafeGraph, and Data Axle is the only option available in ArcGIS Enterprise (Esri documentation, checked August 2026). Worth knowing when you line a "platform" up against a "data provider." Sometimes you are comparing the same data twice.
Buying a movement panel when you needed clean POI records, or the reverse, is the most expensive mistake in this category, and it happens because both get sold as "location data."
Why the published numbers do not settle anything
Record counts are the first number in every deck and the least useful one in the room.
Foursquare's December 2025 open Places release reported 106,205,195 records. An independent count of the same release measured 104,511,073 (Foursquare release notes, checked August 2026). The gap is not scandalous, and it is not the point. The point is that a headline count varies by who is counting and which release they counted, so it cannot carry a purchase decision.
The same holds for the open baseline. Overture Maps Foundation Places counts have been reported at roughly 72 million for a January 2026 release and roughly 75 million in Overture's own July 2026 material. Those are different releases, not contradictions, which is exactly why a count without a release date attached is noise.
What you want instead is a count scoped to your world: how many records exist in the 14 metros you are expanding into, in the categories that matter for your business, with the attributes you plan to filter on. Most vendors can produce that. Very few volunteer it.
The seven criteria
Each row is a question with a checkable answer. Two of them deserve more than a line.
Positional accuracy is a method question, not a quality adjective. Ask whether coordinates are rooftop, parcel centroid, or street centerline. A parcel centroid on a 40-acre power center can sit hundreds of feet from the door, which distorts every drive time, distance band, and trade area built on top of it without ever looking wrong on a map. "Geocoded" is not an answer.
License terms decide what you keep. If you build a scoring model on a vendor's file, some agreements require you to delete derived works when the contract ends. Others do not. That single clause determines whether three years of model calibration belongs to you or to them, and it almost never comes up in the demo.
Freshness has to be measured against real churn
Most POI vendors refresh monthly. The question is whether monthly is fast enough for what you are tracking, and the honest benchmark is how fast the ground actually moves.
The Census Bureau's Business Dynamics Statistics put the US establishment exit rate at 19.3 percent in 2023, roughly matched by entries that same year (Census BDS, checked August 2026). Close to a fifth of the establishments in a national file can be gone within a year. A dataset that lags by two quarters is not slightly stale in a growth market. It is describing a different set of businesses.
Duplicates compound it from the other direction. SafeGraph's own 2024 review describes a single global deduplication pass in May 2024 that removed 325,000 duplicate place records (SafeGraph, checked August 2026). Credit where it is due: a vendor cleaned up and published the number. It is also a measure of how much duplication accumulates in a commercial file between cleanups, which is what inflates competitor counts and distorts saturation math. Our piece on using location data to fix underperforming stores walks through what those distortions do downstream.
So ask for two things: a per-record last-verified date, and the vendor's own measured rate of closed locations still marked open. If neither exists, freshness is a marketing claim.
Movement panels in 2026 are smaller and more regulated
If any part of what you are buying is device-derived, the last three years changed the ground under it.
Opt-in is the ceiling. Apple's App Tracking Transparency prompt has settled at roughly a third of iOS users saying yes, tracked at about 34 percent in mid-2023 and roughly 38 percent by early 2026 (Adjust, checked August 2026). No panel can be more representative than the population willing to be in it.
Enforcement narrowed the rest. The FTC banned X-Mode Social and its successor Outlogic from selling sensitive location data, proposed in January 2024 and finalized that April (FTC, checked August 2026). InMarket Media drew a similar order the same month. Gravy Analytics and Venntel were finalized in January 2025. Kochava reached a proposed settlement in May 2026 after nearly four years of litigation. State law moved with it: Maryland's Online Data Privacy Act took effect October 1, 2025, with enforcement from April 1, 2026, and bans geofencing within 1,750 feet of a healthcare facility.
Consolidation followed the money. Near Intelligence filed Chapter 11 in December 2023, had its plan approved in March 2024, and now operates as Azira.
None of that makes panel data unusable. It does mean provenance is now a procurement question rather than a technical footnote. Ask which upstream sources feed the panel and what the consent basis is for each. "Aggregated from multiple partners" is the answer you should refuse to accept.
Panel bias is the quieter problem. A five-year study of SafeGraph Patterns data published in PLOS ONE found gender and age sampling bias mostly small, but materially larger underrepresentation of Hispanic populations, low-income households, and lower-education groups, varying by geography and urbanization (PLOS ONE, checked August 2026). If your concept indexes toward any of those groups, a national panel is systematically undercounting your customer.
Run the bake-off
Every claim above is testable in an afternoon, because you already own the answer key.
The design detail that makes this work is step 2. Give the vendor the markets, not your stores. If they know which addresses you are checking, you are testing their ability to match a list, not their coverage of a market.
The logic is simple enough to say out loud in a vendor meeting. If a provider cannot reliably find the stores you know are there, its count of the competitors you cannot see is worth nothing.
Open data is now a legitimate control group
The Overture Maps Foundation publishes an open Places dataset monthly, assembled from Meta, Microsoft, Foursquare, PinMeTo, and smaller contributors, with Meta the largest single source. It deliberately excludes OpenStreetMap from Places to avoid the share-alike obligations of the ODbL license, so most records carry permissive terms (Overture documentation, checked August 2026). The foundation reached 50 member organizations in July 2026.
Open data gives you something the category lacked a few years ago: a free control group. Run the same bake-off against Overture first. Whatever a paid provider gives you above that line, in category depth, attribute fill, verification, or coverage in the specific markets you care about, is what you are actually paying for. If a vendor cannot beat a free monthly file in your markets, the price is the whole conversation.
What happens after you trust the file
Picking a provider is the beginning of the work, not the end of it. A clean POI file still has to become a trade area, a competitor set, a cannibalization estimate, and a site decision that a committee will sign.
GrowthFactor works at that layer. POI and foot traffic are inputs, not the deliverable. A site comes back with a score, the lens-level breakdown of what moved it, the trade area, and the competitors and complements around it, then moves into a pipeline your team works from. Cavender's evaluated more than 2,000 sites that way and went from opening 9 new stores in a year to 27, with every new location performing at or better than expected. When Party City's locations hit the auction block, Books-A-Million, the number two book retailer in the US, had roughly 700 sites scored and forecast against its own criteria inside 72 hours.
The data underneath still has to be right. That is why the bake-off comes first. For the wider picture of what else feeds these decisions, see our guides to site selection data and commercial real estate data.
Frequently Asked Questions about Location Data Providers
Common questions from analysts and real estate teams evaluating location data providers.
What should you look for in a location data provider?
Look at seven things: coverage in the specific markets and categories you operate in, positional accuracy of the coordinates, how fast closed locations leave the file, attribute fill rates rather than field counts, the panel basis behind any movement data, the provenance and consent story for the underlying signals, and whether the license lets you keep derived scores and models after the contract ends. A national record count answers none of these.
How do you test the accuracy of a location data provider?
Use your own store list as the answer key. Give the vendor a list of markets, not a list of your stores, and ask for every record in those trade areas. Then score how many of your locations they found, how many they missed, how many duplicates appeared, and how many closed stores are still listed as open. Compare their coordinates against your surveyed ones. Rerun the same query in 90 days to see the real refresh rate.
What is the difference between POI data and foot traffic data?
POI data is a structured record of places: name, category, coordinates, address, and attributes. Foot traffic data is an estimate of visits to those places, derived from panels of mobile devices and extrapolated to the full population. POI data is the map of what exists. Foot traffic data is a modeled estimate of who goes there. They fail in different ways, so evaluate them separately even when one vendor sells both.
Are open location datasets good enough to replace a paid provider?
For a base layer, often yes. The Overture Maps Foundation publishes an open Places dataset on a monthly cadence, assembled from Meta, Microsoft, Foursquare, and other contributors, under permissive licenses. It is a reasonable baseline to test paid vendors against. Where paid providers still earn their price is category depth, attribute fill rates, verification, and support in the specific markets you are expanding into.
How does GrowthFactor compare to Esri Business Analyst for location data?
Esri Business Analyst is a mapping and analysis environment where you choose an underlying POI source, such as Foursquare, Data Axle, or SafeGraph, then build the analysis yourself. GrowthFactor treats location data as an input rather than the deliverable: the trade area, competitor set, demographics, and site score arrive together with the inputs that moved them, and the site moves into a deal pipeline your team works from. Cavender's evaluated more than 2,000 sites that way while going from 9 new stores in a year to 27.