Methodology
How Appdex estimates are made
Every figure on Appdex comes out of a purpose-built estimation engine that never stops running β and we treat its output honestly. No fake precision, no black-box certainty: each number ships with a confidence level, and figures verified against real disclosures are labeled as such.
The pipeline, at a glance
Hover or tap a stage. The what is public β the how is the product.
Collectors sweep public App Store signals around the clock, across 16 storefronts.
Charts, ratings behavior, pricing and catalog movement β dozens of public signals, captured continuously. Nothing private, nothing scraped from behind a login.
The engine blends those signals into one estimate per app β millions of rows re-processed every day.
How the signals are weighted, which ones dominate for which kind of app, and what corrects what β that blend is Appdex's trade secret, and it changes as the engine learns.
Every night the engine sits an exam it cannot cheat: tens of thousands of independently known figures.
Estimates are compared against real-world figures the model has never seen. Changes that don't improve real accuracy don't ship β however clever they look.
Only what passes ships β each number with its confidence label.
Verified means a real disclosure. Calibrated means matched to live market data. Everything else says exactly how sure the engine is β no fake precision.
Where the data comes from
Three classes of input, in decreasing order of authority: verified disclosures (figures developers and companies have published themselves β these anchor everything and are labeled verified), public store signals (dozens of them, collected continuously β we don't enumerate which ones carry the most weight), and market context (cross-store and category-level behavior that keeps individual estimates consistent with how the market actually moves).
How estimates are validated
Against held-out ground truth the engine has never seen β tens of thousands of apps with independently known revenue and download figures. The exam re-runs continuously; publish-facing work is gated on it. Category-level accuracy differs, and that difference is reflected in the confidence level each page displays. This is what honest confidence means: we don't ask you to trust the engine β we test it every day and show you the grade.
Update cadence
The brains never sleep. Estimates rebuild nightly; ranking signals refresh through the day; collectors run around the clock; verified disclosures are folded in as they appear. The date on each page reflects the data it was built from.
What we don't do
We don't sell or export the underlying dataset. We don't show four-significant-digit "precision" the evidence can't support. We don't hide when the honest answer is a wide range or "low confidence". And we don't reveal the blend β see above.
Who runs Appdex
Appdex is independently built and operated. Questions, corrections, or a disclosure you'd like reflected? Contact us β developer-provided figures are verified and folded in, and they make the engine sharper for everyone.
See it in practice: Insights Β· What a confidence band means
Start 3-day free trial β