You should stop optimizing for installs and switch from CPI to ROAS as soon as you have a stable, measurable revenue signal within your payback window. For most apps that means: a clear D1–D7 monetization event, enough daily conversions per geo, and an LTV vs. CAC model you actually trust.
Staying on CPI after that point means you’re paying for users who look cheap, not users who pay back. The switch is less about campaign structure and more about deciding which revenue signals are strong enough to let the algorithm trade off volume for profit.
Why is optimizing only for installs a problem?
CPI optimization solves an early-stage problem: get users into the app at a reasonable cost. It doesn’t solve the business problem: grow profitable revenue.
When you stay on CPI too long:
- You bias towards low-quality inventory and geos that drive cheap installs.
- You attract users who open once, never monetize, and still count as a “win”.
- Algorithms learn the wrong pattern: “people who install easily”, not “people who generate revenue”.
Most major networks (Meta, Google, ASA, TikTok) now support value- or ROAS-optimized campaigns. Those bidding modes are designed to fix exactly this: they bid more for users whose predicted LTV > target CAC, even if the install itself costs more.
When is it too early to switch from CPI to ROAS?
Switching to ROAS bidding too early just hands the algorithm noisy, sparse data. That usually means:
- Wild CPI swings
- Limited scale
- Optimizing to random whales instead of a repeatable profile
It’s usually too early to switch from CPI to ROAS if:
- Your main revenue events happen later than D7–D14 and you don’t have a good proxy event.
- You’re getting fewer than ~50–100 post-install revenue events per campaign per week on a major network.
- Your MMP or SKAN schema can’t reliably attribute in-app revenue yet.
In these cases, staying on CPI or optimizing to a cheaper upper-funnel event (e.g. registration, tutorial complete, first purchase) is often smarter while you build data density. As AppsFlyer and Adjust often point out, conversion volume is the fuel for any optimization algorithm — starve it and performance degrades.
What signals tell you it’s time to switch to ROAS?
You’re ready to move from CPI to ROAS when three conditions are true:
-
You know your payback math
- You have a target D7 / D30 ROAS (e.g. 20–30% at D7, 80–100% at D30) based on cohort analysis.
- You understand blended vs. channel-level ROAS and can monitor both.
-
You have early revenue or strong proxy events
- Direct revenue: IAP, subscriptions, or ad revenue that shows up in D0–D3.
- Proxies: account creation + KYC for fintech, level X reached for games, add-to-cart or trial start for subscriptions.
-
You have enough volume per optimization event
- At least tens of purchases or high-intent events per day at the account level, and dozens per week per campaign.
Once those are in place, keeping campaigns on CPI is just an opportunity cost.
How do you phase the switch from CPI to ROAS?
Don’t flip everything overnight. Phase the transition to control risk and keep learnings clean.
Step 1: Lock in your measurement
- Ensure your MMP events, values, and revenue are flowing correctly across all networks.
- For SKAN/AdAttributionKit, map conversion values to early monetization or meaningful proxy events, not just installs.
- Align BI and UA on LTV windows and ROAS definitions so you’re not optimizing against moving targets.
Step 2: Move from installs to revenue-proxy events
- Shift campaign optimization from “install” to a lower-funnel event (registration, tutorial complete, first purchase) while still bidding on CPA.
- Use this phase to verify: does cheaper CPI actually mean cheaper CPA and better cohort ROAS? If not, your targeting or creatives are misaligned.
Step 3: Test ROAS/ value-based campaigns in parallel
- Duplicate top CPI/proxy campaigns and run new ones optimized for value/ROAS with conservative budgets.
- Start with a modest target ROAS and a payback window you can realistically hit (e.g. D7 for casual games, D30+ for some subscriptions).
- Compare on profit, not just CPI or install volume.
Step 4: Gradually re-allocate budget
- As soon as ROAS-optimized campaigns show more stable cohorts (even at higher CPI), shift budget from CPI to ROAS.
- Kill CPI campaigns that clearly underperform on real revenue, even if on-paper CPIs look attractive.
How does the switch from CPI to ROAS affect your campaigns?
Expect the following changes when you move to ROAS optimization:
- Higher CPIs, higher value: CPI often rises 20–50% while ARPU and ROAS improve. That’s the trade you want.
- Slower learning: ROAS-optimized campaigns need more time and budget to exit learning phases.
- Creative feedback loop changes: Winning creatives are those that attract payers, not just clickers.
Here’s a simplified comparison of staying on CPI vs. switching to ROAS:
| Dimension | CPI Optimization | ROAS / Value Optimization |
|---|---|---|
| Primary goal | Lowest cost per install | Highest revenue per dollar spent |
| Core metric | CPI, installs | ROAS, ARPU, LTV:CAC |
| Ideal stage | Pre-launch / early launch | Post-product–market fit, stable monetization |
| Data needed | Click → install | Post-install revenue or strong proxy events |
| Typical outcome | Volume, mixed quality | Less volume, higher-value users |
| Risk if overused | Cheap but unprofitable users | Slow scale if targets are too aggressive |
What mistakes should you avoid when switching to ROAS?
Common pitfalls when moving from CPI to ROAS:
- Setting targets based on wishful thinking: Target ROAS must come from real cohorts, not board slides.
- Ignoring post-install product issues: No bidding strategy fixes a broken onboarding or paywall.
- Over-fragmenting campaigns: Too many ROAS campaigns split conversion volume, starving algorithms.
- Judging too early: Killing value-based campaigns after 2–3 days ignores how long it takes to get clean ROAS data.
The cure is simple: fewer campaigns, clear targets, and decisions made on cohort ROAS and payback, not daily CPI charts.
FAQ
When is the earliest I should consider switching from CPI to ROAS?
As soon as you have a consistent early revenue or proxy event and at least dozens of those events per week per campaign. For many apps that’s a few weeks after global launch, once monetization stabilizes.
Should I ever stay on CPI long-term?
Only if you truly can’t measure revenue or high-intent events, or if the channel doesn’t support value optimization. Even then, you should benchmark CPI against cohort-level ROAS to avoid scaling unprofitable traffic.
What if my app’s revenue comes mostly after D30?
Use strong proxy events that occur earlier (e.g. onboarding completed, trial started, level X reached) and optimize to those. Then cross-check that cohorts acquired with those proxies still hit long-term payback.
Do I need perfect LTV predictions to run ROAS campaigns?
No. You need directional accuracy and stable patterns, not perfect forecasts. Focus on short- to mid-term windows (D7–D30) where you have enough historical data.
How do I explain higher CPIs to stakeholders when we switch to ROAS?
Reframe reporting around profit and payback: highlight ARPU, ROAS, and LTV:CAC. Show side-by-side cohorts from CPI vs. ROAS campaigns to prove that higher CPI can still mean higher margin and faster payback.