How to estimate App Store revenue (and what the estimate is worth)
7 min read · updated
Every App Store revenue number in public circulation is the end of a chain of estimates, and each link in that chain can be wrong by a multiple. This is the chain, so you can decide for yourself how much weight the output carries.
The chain, link by link
- Ratings are counted, not estimated. Apple publishes each app's lifetime rating count and its current-version rating count. This is the one measured number in the whole chain, and everything downstream leans on it.
- New ratings a month is a difference, so it needs history. Apple exposes today's count and nothing else. If nobody recorded yesterday's, there is no velocity to compute. This is why chart-history tools only have data from the day they started collecting, ours included.
- Installs come from a ratio, and the ratio is the weak link. Roughly 75 installs per rating in the US, before any correction. That is an order of magnitude, not a measurement.
- Revenue needs a conversion rate and a price you cannot see. Apple does not publish subscription take-up, churn, or how much of the price a promotional tier gives away. This link is the one that should make you stop treating the output as a number and start treating it as a band.
The correction almost everyone skips
Users in different storefronts rate at wildly different rates, so the same install count produces very different rating counts depending on where it happened. Correcting for it changes the answer by more than a factor of four:
| Storefront | Rating propensity vs US | Installs behind one rating |
|---|---|---|
| United States | 1.00 | ~75 |
| United Kingdom | 0.85 | ~88 |
| Germany | 0.60 | ~125 |
| South Korea | 0.55 | ~136 |
| Japan | 0.45 | ~167 |
| Turkey | 2.00 | ~38 |
Read the two ends of that table together. An app with 10,000 Turkish ratings and an app with 10,000 Japanese ratings look identical in any tool that counts ratings, and the Japanese one is plausibly four times the size. Turkey is not a rich market that punches above its weight; Turkish users simply rate more.
What to do with an estimate instead of believing it
Use it as a comparison, not a forecast. The ratio between two apps in the same storefront and category is far more stable than either absolute number, because the shakiest links in the chain apply equally to both and cancel.
Then sanity-check against the entry wall rather than the market size. A category doing tens of millions of modelled installs a month behind a median top-grossing app with 874,000 ratings is not an opportunity you can size your way into. Finding a profitable niche covers reading that wall.
How to spot an estimate pretending to be a measurement
- A single figure with no band. Real modelling produces a range, and a wide one.
- Revenue quoted to the dollar. The conversion rate underneath it is not knowable.
- Cross-country comparisons with no propensity correction. See the table above.
- Any claim of "actual" downloads. Apple does not publish them. There is no actual.
We label our own install and revenue figures as estimated everywhere they appear, on the method pages and inside every run, for exactly this reason. The scoring methodology names each input and where it came from.
Questions
- Can you find out how much an iOS app really earns?
- No. Apple publishes no downloads and no revenue, to anyone, and no third party has a feed of them either. Every figure you have ever seen from an ASO tool is modelled, usually from panel data or from rating counts. The useful question is not what an app earns, it is whether the ratio between two apps is stable enough to compare them, and it usually is.
- How many downloads is one rating worth?
- Roughly 75 in the US, as an order-of-magnitude anchor rather than a constant. It is not a fixed number: it moves with category, with how hard an app nags for ratings, and hardest of all with the storefront. Comparing rating counts across countries without correcting for that is the most common way an estimate goes badly wrong.
- Why do two tools give completely different revenue estimates for the same app?
- Because they are different models, not different measurements. One may be panel-derived and another rating-derived, and neither has ground truth to calibrate against. Disagreement between them is expected and tells you nothing about which is right.