$4 CPI on iOS. 27% D0 ROAS. 10% D1 conversion. Week 1 on Meta, a casual game, and everything looked incredible. We still call it the golden cohort.
Here is what we saw looking back.
Three inputs, not one
Three inputs. Over 14 weeks all three moved, sometimes in opposite directions, and the same D0 was reached twice through completely different user profiles before it declined.
D0 ROAS is a ratio with three moving parts, and reading it as a single number hides which part moved.
- Weeks 1 and 2: $4 to $6 CPI, 10% conversion, $10 AOV, D0 at 23%. CPI carried everything. AOV was modest and did not need to be more than that.
- Weeks 6 and 7, after a store featuring: $10 CPI, 8% conversion, $29 AOV, D0 at 22%. CPI doubled but AOV tripled.
That was the surprising part. The same D0 above 20% came from two completely different user profiles. One subsidised by cheap installs, the other by high-value purchasers.
Then the input that mattered moved
AOV started falling. $29, then $25, then $23, then $18, over four weeks. Same campaign, same setup, same optimization objective.
D0 dropped from 22% to 11%.
Decomposing the drop:
| Input | Share of the drop |
|---|---|
| AOV decline | 62% |
| Conversion drop | 38% |
| CPI increase | 11% |
Value per buyer was responsible for the damage; CPI barely moved the needle.
Why this is hard to diagnose on iOS
When you run value optimization on iOS, Meta uses D0 revenue as its targeting signal. If that revenue changes for product-side reasons, offers, eCPMs, engagement, then Meta’s targeting shifts with it.
It is a feedback loop. The revenue is the signal.
Meanwhile the game-wide numbers said this was not only an acquisition problem. ARPDAU dropped 30%, IAP per user was down 33%, ad revenue per user down 25%. Revenue per active user was softening across the entire player base.
The revenue mix had shifted too. Week 1 was 85% IAP. By week 6 it was an even split between IAP and ad revenue, which meant AOV now depended on ad eCPMs as much as on purchase behaviour.
The one number that held
Cost per purchase. While CPI, conversion, and AOV all moved independently, CPP captured both acquisition cost and conversion in a single figure.
Every week CPP came in under £30, D0 landed above 20%. Every week it did not, D0 missed.
If you track one number daily, that is a strong candidate. Just do not expect it to tell you which of the three inputs moved, because that is what the decomposition is for.