The Attribution Chain Most Coverage Skips

Every headline announcing that a particular app topped a billion downloads, or that the App Store generated a specific sum last quarter, traces back to the same small cluster of firms: Sensor Tower ↗, Appfigures, and data.ai (formerly App Annie). None of them receives raw transactional data from Apple or Google. What they sell are estimates — constructed from panel data, store-scraping, and proprietary modelling — and understanding the construction is the first step to reading the numbers honestly.

The core methodology across all three firms combines a panel of opted-in users whose install and purchase behaviour is observed directly with algorithmic extrapolation to the broader market. Panel size, geographic coverage, and the weighting model are the variables that diverge. Sensor Tower, after its 2024 acquisition of data.ai, now operates the largest combined panel, though the two platforms continue to publish separately. Appfigures draws more heavily on App Store Connect ↗ data volunteered by developers themselves, which gives it stronger ground-truth anchoring for download counts but thinner coverage of in-app revenue.

A laptop screen showing a developer analytics console with revenue figures visible, an adult's hands resting on the keyboard, photographed in office light
A developer analytics console. Almost every public figure here is an estimate read off a screen.Photo: Jakub Zerdzicki / Pexels

The commissioning layer is where headlines are made and where the attribution chain becomes opaque. A streaming service wanting to announce record downloads will commission a report from one of these firms, then issue a press release citing "data from Sensor Tower" or "according to data.ai." The report's methodology page, the confidence interval, and the commissioning party's identity are rarely reproduced. Journalists citing the estimate inherit the uncertainty without inheriting the caveat. When the Korea Communications Commission cited download figures during its 2021 in-app purchase legislation hearings, or when the Epic v. Google trial admitted market-share evidence in the US District Court for the Northern District of California, the originating estimates passed through several hands before reaching the record.

The firms' own disclosures note that revenue estimates for markets with lower ARPU — India, Southeast Asia, Brazil — carry wider error bands than US or Japanese figures, because panel density is thinner where smartphone penetration grew fastest. That divergence matters when regulators in New Delhi or Brasília cite app-economy data to justify intervention: the numbers least likely to be accurate are the ones likeliest to underpin emerging-market policy decisions.

How the estimates are built

Panel dataa pool of opted-in users whose app installs and purchases are observed directly, then extrapolated to the full market
Store-scrapingautomated collection of publicly visible App Store and Google Play metadata (rankings, review counts, pricing)
Developer-submitted datasome firms (notably Appfigures) incorporate data volunteered by developers via platform dashboards
Proprietary weighting modelthe algorithm that scales panel behaviour to market totals; not publicly disclosed by any firm

What the data is genuinely good at is directional movement over time within a single platform. Whether an app's downloads rose or fell quarter-over-quarter, which category accelerated after a rule change — these relative signals are more reliable than any absolute figure. The absolute figures are best understood as the industry's agreed-upon approximations, not audited facts.