Digital advertising still runs on a television map

1 · Broadcast bought a standard, and it worked

Television faced this problem first. It needed one agreed geography so that a spot bought in one place could be counted in the same place, and it solved the problem commercially rather than technically: Nielsen defined the Designated Market Area, registered DMA as a trademark in 1979, and licensed the map. Comscore now sells a parallel definition. Both cut the United States into 210 markets assembled from whole counties, and both describe television viewing behaviour.

The arrangement worked because everyone rented the same map. Standardization by subscription is still standardization, provided nobody changes vendor.

2 · Digital rents that map, from whichever vendor

On June 22, 2026 Meta replaced Nielsen DMAs with Comscore Markets in its automotive model ads, citing the want for a partner solution that is more sustainable and scalable for long-term measurement. One ad product so far, but the mechanics are what matter: the swap changed the supplier and kept the paradigm. New York moved from DMA 501 to Market ID 2001, boundaries shifted at the edges, and every automotive campaign still keyed to the old identifiers stopped delivering until it was remapped.

Google never opted out either. Google Ads sells United States location targeting by Nielsen DMA region, listed in its own documentation alongside country, state and postal code. YouTube TV sells geography at state, DMA and ZIP. Display & Video 360, Campaign Manager 360 and Google Ad Manager all carry the same unit.

3 · The sellers do not measure on it either

The more revealing fact is what happens when the same companies measure rather than target.

Google’s marketing mix model fixes no geography at all. Meridian takes user-supplied geos, described in its documentation as mutually exclusive regions such as states, cities, DMAs, or even multiple countries. That documentation then warns that “different geo aggregation grouping methods can lead to different MMM results”, and advises that “it may be better to fit a model at a finer geo granularity and exclude the smallest geos, rather than aggregating geos to a coarser level”.

Google’s geo-experiment literature goes further. The methodology does run on DMAs, and Google’s own researchers name the resulting constraint: the number of available geos is limited, roughly 210 in the United States, which caps statistical power and limits how many experiments can run at once. Two subsequent bodies of work exist largely to compensate. Time-Based Regression was developed because matched-market tests have too few geos to regress across. Supergeo Design exists to construct better experimental units when the available geos are too few.

4 · Three bars a unit built for digital has to clear

1. Open.Nobody licenses the map, and nobody can renumber it. A measurement unit whose identifiers are somebody’s intellectual property inherits that party’s commercial decisions.
2. Fixed. A boundary cannot move underneath a running test. Administrative geography moves constantly: postal boundaries are redrawn every decade, markets are re-cut, and any time series built on them silently compares two different places. The boundary-drift section of this explainer animates exactly that.
3. Correctly grained.Finer than a television market, because a market that swallows a whole metro cannot separate one trade area from the next town over, and because Google’s own guidance prefers finer granularity to coarser aggregation. Coarse enough to stay outside the precise-geolocation line that United States state privacy law draws at a radius of 1,750 ft, and 1,850 ft in California. That line falls between resolution 7 and resolution 8, which is why grain here is a compliance question and not only an accuracy one.

5 · What this explainer demonstrates

The remaining sections are the working proof rather than the argument. The grid is a global hexagonal partition: every point on Earth in exactly one cell, sixteen nested resolutions, no borders in the geometry, no licence. Any administrative boundary resolves to a weighted set of cells and back again losslessly, which is what lets postal-keyed sales panels and CRM files join to it without being reformatted. Activation rolls the signal grain up to whatever footprint a platform can actually execute. The composite unit crosses cell with ISO week and hour, because a geography without a clock cannot separate a Tuesday morning from a Saturday night in the same place.

None of it requires replacing what already runs. Every conversion on this page is open source and runs in a standard warehouse today. Platforms keep their native encodings, and the unit converts in and out.

6 · Test it on your own numbers

An argument about measurement is only worth as much as it survives contact with a real ledger. If you want to check this one, tell us the geography you care about and the question you are trying to answer, and we will scope what it takes to run it against your own data.