How accurate is AppInsider?
Every published figure we can find, next to what our model says for the same app. Across 69 figures, our model is typically off by 1.58× (1.00× would be exact) and within 1.5× on 28 of 69. Checked 2026-11-11, every day.
The honest number: one app in five is held out: its published figures are never used to tune anything. On those 23 figures our model is typically off by 1.57×, within 1.5× on 10. That is the error to expect for an app nobody has published figures for.
How sure we are: every estimate shows a range and a confidence, the chance the real number is within half to double ours. The ranges are set from the errors on this page, not from the models' own assumptions. Across 361,194 estimates on the site, the average confidence is 73%. Honesty check: on the 13 figures we never tune on, we would have claimed 79% confidence, and 85% actually landed within 2×.
Where we can measure (Google Play install counts, Tencent MyApp in China) we show measured numbers. Everything else is an estimate, labelled as one. Where a company or tracker publishes a figure, the site pulls the app toward it; this page shows our model before that pull, so it stays an honest test. Rows marked "calibrated" are figures the model is also tuned on, which flatters it. The published figures are for one past period and ours is today's rate, so part of any gap is real change.
Downloads in a month
both stores, worldwide excl. China Android. × = ours ÷ published.
Monthly active users
on mobile. × = ours ÷ published.
Revenue in a month, after store fees
both stores. × = ours ÷ published.
Yearly consumer spend
from company results (a quarter x4 where noted). × = ours ÷ published.