Why cohort ROAS beats install metrics for subscription apps

If you run Apple Search Ads for a subscription app, the dashboard is lying to you — not on purpose, but by omission. It shows impressions, taps, and installs. None of those tell you the only thing that matters: did this keyword produce a paying subscriber, and was it worth it?

The problem with cost-per-install

Cost-per-install (CPI) treats every install as equal. But for a subscription app, an install is just the start of a free trial. Two keywords can have identical CPIs and wildly different economics:

• Keyword A — $2 CPI, 8% of trials convert to paid.
• Keyword B — $2 CPI, 55% of trials convert to paid.

On a CPI dashboard these look the same. In reality, Keyword B is ~7× more valuable. Optimize on installs and you’ll happily scale the one that never pays back.

What cohort ROAS measures instead

Cohort ROAS ties ad spend to the revenue the users you acquired actually generated, grouped by when they installed. You stop asking “what did a tap cost?” and start asking “for every dollar I spent on this keyword last month, how many dollars came back?”

To compute it you need two things joined together:

1. Spend, per keyword — from Apple Search Ads.
2. Revenue, per acquired user — from your subscription data (e.g. RevenueCat), attributed back to the keyword via Apple’s AdServices token.

With those linked, every keyword carries trials, attributed revenue, and a real ROAS — not a proxy.

Why your fresh cohorts will look terrible (and that’s fine)

The catch: most installs begin as free trials, so a cohort you acquired this week shows trials but almost no revenue. The money lands later, when those trials convert. A brand-new keyword can look like a 0.0× disaster and still be your best performer once the cohort matures.

So read new keywords in this order:

1. Trials — is it producing trial starts at all?
2. Trial-conversion rate — are those trials the kind that convert?
3. Cohort ROAS and payback — once the cohort is 30–60 days old.

Judging a two-week-old cohort on revenue is the single most common way teams kill their best keywords early.

The lever almost nobody uses: payback window

Once you measure cohorts, a new lever appears. If you only count revenue in the first 30 days, a keyword has to pay back fast to look profitable. Extend your measurement to 12 months and keywords that looked unprofitable start clearing the bar — because subscriptions compound.

A 20% annual return ($1 → $1.20) is excellent, but it takes patience most small operators don’t have. That patience is exactly the edge: accept a longer time-to-recover and the pool of keywords and geos you can profitably bid on expands dramatically.

Build the LTV model first (3–6 months of cohort data), extend the window incrementally, set a return floor, and test on one geo before the whole account.

Where to start

You don’t need a data team. You need spend and revenue in the same place, attribution wired through AdServices, and the discipline to judge cohorts on trials first and revenue later.

That’s exactly what TapPilot does — it connects your Apple Search Ads spend to RevenueCat subscribers and gives you cohort ROAS, payback curves, and trials per keyword out of the box. See how it works.

Once you have those numbers, use them to calculate a CPT ceiling for each keyword so every bid in your account is grounded in the same subscriber economics.