I do not quote industry averages, because I cannot verify them. What follows is what accounts I have run actually did, each row linked to the case or the post it comes from. Use them as reference points rather than targets: every one depends on the product behind it.
This page is for someone with their own numbers in hand, asking whether they are good. Every row has a source you can open. If you want the reasoning behind a row, the source is where it lives; if you want your own numbers read, that is the audit.
Return on ad spend
| Metric | From | To | Where | Source |
|---|---|---|---|---|
| D0 ROAS | 20% | 43% weekly record | Videa AI and Cleaner Pure, subscription, spend past $200K a month | source |
| D0 ROAS | 8 to 12% | 17 to 23% | Horse Racing Solitaire, game, soft launch to scale over two years | source |
| Blended ROAS | 70% | 230% | EatBetter, subscription, in two months | source |
| ROAS by D28 | 100% | Honig Games, US cohorts, per the client | source | |
| Channel D28 ROAS | TikTok below parity | Meta 1.61x | One subscription app across Meta, Apple Search Ads, Google and TikTok | source |
Cost per install
| Metric | From | To | Where | Source |
|---|---|---|---|---|
| Blended CPI | £9.43 | £5.38 | Horse Racing Solitaire, in one month | source |
| Creative test, best ad | £0.65 | 175 ads over four months, 14 under £1.00 | source | |
| Creative test, blended | £1.22 | Same account, same period, all 175 ads | source | |
| UGC ad, product tried on camera | CPI down 40% | Diress, against the account baseline | source | |
| UGC talking head | CPI down 22% | A horse racing game, against the account baseline | source |
Revenue, and the cost of average
How to read these
Every row is one account, one period, one product. The ROAS a subscription app can reach depends on retention and price before it depends on any campaign, and a game with a live economy can look worse on day zero and better on day 28 than an app whose revenue arrives up front. The tests behind the bigger claims are written up in full: a $606K test on whether day zero value or cost per purchase predicts D28 ROAS, a $100K test of Google bid strategies, and a $383K audit of how often Meta hits a tROAS target.
Questions people ask
What is a good CPI for a subscription app?
The one your payback can afford. CPI on its own tells you almost nothing; a campaign can bring 10,000 cheap installs and zero subscribers. I set the CPA ceiling from unit economics first and judge the rest on revenue.
What ROAS should I target on day zero?
It depends on how fast your revenue arrives. The two accounts above sit between 17 and 43 percent on day zero and both pay back. What matters is the curve from day zero to day 28, and the post on day zero value versus CPA shows which one predicts it.
Why no industry averages?
Because the ones in circulation are unsourced or reported by the vendors themselves, and I would be repeating a number I cannot check. Every figure here has a case behind it that you can read.