If you’re running Meta Ads for a subscription app and still using the same playbook from last year, the bad news is that it’s probably not working anymore.
Andromeda, Meta’s ads retrieval engine, changed how Meta’s algorithm processes signals, finds audiences, and optimizes delivery. The old “spray and pray” approach is dead. The new game is about signal quality, creative velocity, and knowing exactly when to trust automation, and when to override it.
I recently did two detailed interviews on this topic with Adapty, and I want to share the full playbook here. The questions came from the hosts, myself and the comments we’ve collected. This is what I’m actually doing with clients right now, not theory.
Start with Meta. Period.
I get asked constantly: “Should I start with Google? TikTok? AppLovin?”
I always answer that if you can’t make Meta work, another channel rarely saves you.
Meta comes first for three reasons:
- Meta has the largest, most diverse audience
- The feedback loop is fastest (you can validate creatives in 1 to 3 days)
- The algorithm is the most sophisticated at finding your buyers
My order: Meta, then Google, then Apple Search Ads, then TikTok.
| Channel | When it earns a place | How it takes a budget increase |
|---|---|---|
| Meta | First, before any other channel | Scale freely |
| Google Ads | It can work, but it’s harder, especially for iOS. On both OS it needs more budget and takes about 2 weeks to learn | Scale slower, in smaller steps. More sensitive |
| Apple Search Ads | Nice to have, but competitive keywords are expensive and niche keywords often have no volume | No special care needed |
| TikTok | Only for a Gen Z app that can go viral there, and only if you’re willing to go all in. Otherwise it’s a distraction, whatever the loud voices say | Needs care, like Google |
| AppLovin | Not ranked. AppLovin requires MMP tracking set up before a campaign goes live and recommends a daily budget that drives at least 15 to 20 conversions a day across the campaign, so I bring it in once Meta is working and monthly spend sits well into six figures. How I would test whether it is incremental is its own article | Not covered in this playbook |
When I compared day 28 ROAS across all four channels on one subscription app, Meta led clearly.
And the rule that will save you money is to never optimize for installs if you want revenue. You can test creatives on install campaigns, but your real campaigns must optimize for trials or purchases. Install optimization brings you people who install apps, while purchase optimization brings you people who buy things, and those aren’t the same people.
Where to test: the Tier 1 West strategy
A lot of teams test in cheap countries first, India, Brazil, Southeast Asia, thinking they’ll “validate” before spending real money in the US.
This usually backfires.
Creative performance varies by culture, economy, and pricing tolerance. What works in Tier 2/3 often misleads you about what will work in Tier 1. You end up optimizing for the wrong audience.
My approach is to test where you plan to scale.
One cheap round before that is fine. The elimination round runs in a single English speaking Tier 2 country, and its only job is to pick the best version of each concept. Prove whatever survives in the market you plan to scale in.
I run creative tests in what I call “Tier 1 West+”:
- English as the main language: US, Canada, UK, Ireland, Australia, New Zealand
- Other languages: Germany, France, Italy, Spain, Netherlands, Switzerland, Nordics, Hong Kong, Japan, Korea. Set language targeting to English in the countries where English is not the main language.
Why? This reaches:
- People who speak English at native level
- Expats living in those countries
- Locals who use apps in English
This lets you bundle all Tier 1 West into a single campaign while keeping CPIs reasonable. Once you have winners, you can split by language and localize.
Don’t run from expensive geos. That’s where the subscription money is.
The metrics that actually matter
Everyone obsesses over CPI. But CPI is a vanity metric for subscription apps.
For creative testing (install campaigns): the metric that matters is IPM (installs per 1,000 impressions). And the amount spent.
I use this benchmark for US campaigns optimized for installs:
| IPM (US, install campaigns) | What it means |
|---|---|
| Below 2 | Your creatives aren't ready. Keep testing. |
| 2 to 5 | You're ready to scale or explore serious campaigns. |
| 5 to 10 | Strong. Scale it and watch whether it holds. |
| 10+ | You have a winner. Push it hard. |
Creative validation is cheap relative to the account. $20 to $30 a day per test cell answers the IPM question, even when the account behind it spends six figures a month.
For subscription and trial campaigns: ROAS is king. But you also need signal volume. Aim for at least 5 to 10 optimization events per day per campaign. Meta’s commonly cited guideline is about 50 per ad set a week. Below that, learning is slow and unstable. A common way to fall short is spreading one budget across too many campaigns.
Spend shows what Meta chose to deliver. It says nothing about what those users were worth. On one account, the five ads Meta spent most on all missed target, and removing them helped.
I’ve seen people pause their best creatives because the CPI was $0.80 instead of $0.50. The $0.50 creative brought cheap installs that never converted. The $0.80 creative brought buyers.
Cheap installs mean nothing without performing the same over days and weeks.
Campaign structure: the three phases
I use this structure.
Phase 1: Creative Lab (Android install campaigns). Start on Android because it’s cheaper and better attribution. Run campaigns optimized for installs with broad targeting. Your only goal is to find creatives with strong IPM.
Test a lot. I mean it. Test 100, 500, even 1000 ads if you can produce them. Out of 1000 ads, maybe 10 will be winners. Test different concepts, and from the winning concepts, different variants. Don’t slack off, do the homework.
Let each ad spend at least $1 so it gets some delivery. Use rules to pause ads that hog all the budget, you want distribution across creatives.
Phase 2: Validate winners on iOS. Take your Android winners (the top 10 or so by IPM) and test them on iOS install campaigns. Some will carry over, some won’t. People are people, but platforms have quirks.
Maybe 3 out of 10 survive as true winners.
Phase 3: Conversion campaigns. Take those survivors and put them into trial and subscription campaigns targeting Tier 1 West with English language.
Use your custom optimization event, something like “trial started AND not canceled within 2 hours” or a combination of trial + direct purchase.
Keep targeting broad. Only narrow by gender or age if your product clearly calls for it (women’s health app: target women). Otherwise, let the creatives find the audience. You can also “suggest” an audience on the ad set level if you are not sure.
Phase 4 (later): language splits. Once you have volume, split into language groups: a bundle for English speakers, one for German speakers, one for French speakers, and so on.
CapCut makes video translation ridiculously easy now. Don’t shy away from translating your winners into German, French, Spanish. It’s low effort and can add significant scale. On one account, language explained 4.8% of the variation in ROAS and the concept 25.2%, so test in English and localize what wins.
The Advantage+ question
“Should I use Advantage+ campaigns?”
This is the most complicated question in Meta Ads right now, and my answer depends on your monetization model.
- How does your app make money?
- Direct purchases, no trial: use Advantage+. It works great here because Meta sees clear, immediate conversion signals and optimizes effectively, so I trust automation from day one.
- Free trials: be careful, and go to question 2. Advantage+ tends to overoptimize for cheap trial users, especially younger demographics who trial everything but rarely convert.
- Is your trial data stable yet?
- Not yet: my rule is no Advantage+. I use manual setup so I can control signals and filter by country, age and gender as needed.
- Yes, with a working purchase flow and stable conversion data: then you can test Advantage+.
I’ve seen that overoptimization kill campaigns. Meta proudly reports tons of trial starts, but none of them become paying subscribers.
Either way, you don’t need custom audiences. Let creatives do the targeting work.
Signals: what to send Meta
Meta’s algorithm learns from your events. But not all events are created equal.
Events that DON’T work well as optimization targets:
- “Onboarding completed”: too easy, most users complete it. Not tied to money.
- “Purchase started”: doesn’t guarantee money, as most will not purchase from here.
Events that DO work:
- “Trial started AND not canceled within 2 hours”
- Combined event: trial not canceled + direct subscription purchase
The 2 hour filter works because cancellations come early. In RevenueCat’s 2026 data, 55.4% of cancellations in three day trials happen on day 0 (State of Subscription Apps 2026).
The trick is finding an event that’s:
- Close enough to revenue to attract quality users
- Frequent enough for the algorithm to learn (aim for 5+ per day per campaign)
I get this question a lot: “My organic ratio of installs to trials is only 2%. Is that too low?”
Don’t worry. Paid optimization can push that ratio to 10 to 15%. When you tell Meta to optimize for trials, it finds people who actually start trials. Your organic ratio isn’t your paid ratio.
Just make sure your ratio of installs to events is at least 5%. Below that, you’ll burn too much money getting enough events for learning.
How long to test
For creative testing: fast. Let each ad spend at least $1 (use rules for this). After about 10 installs in an ad set, you can see which creative drove them and at what cost. I sometimes use automation rules to pause ads after $1 spend so one winner doesn’t eat the whole budget.
For trial and subscription campaigns: 3 days, few creatives, few hundred dollars.
Strong campaigns work fast. You should see your first trials or purchases within 1 to 3 days. If you’ve spent $200 to $300 with decent metrics at the top of the funnel (good CPM, CTR, IPM) but zero trials or purchases, stop.
Don’t put too much trust in the “learning phase.” In my experience, weak campaigns don’t magically fix themselves after day 5 or the 50th event. If it’s not working in 3 days, something upstream is broken.
Check:
- Creatives (is IPM below 2?)
- Paywall (is your offer clear?)
- Tracking (are events firing correctly?)
- The product itself (is there actual demand?)
Try a “Tier 0” test: US, CA, AU, DE, FR with English language only. Spend meaningfully for 48 to 72 hours. If still nothing, pause and regroup.
Scaling without breaking things
“If I increase my budget, will I tank my campaign?”
Ask 10 media buyers, get 10 different answers.
My answer is that Meta can take it. I double budgets comfortably.
The “learning phase” is overhyped. If your campaign is solid, it stabilizes again fast. Minor changes (copy tweaks, asset swaps) rarely tank performance.
My preferred way to scale is to duplicate the best campaign and raise the copy’s budget. Both compete for the same people, so watch the pair together. If the duplicate works, great, and if it doesn’t, your original stays untouched.
Each channel takes a budget increase differently, and the channel table at the top shows how much care each one needs.
Remember that performance fluctuates naturally, daily, weekly, even monthly. If results drop after a budget change, it might just be normal volatility, not your edit.
Subscription economics: the cash flow reality
The unique challenge for subscription apps is long payback periods.
If your payback is 6 months, you’re funding 6 months of growth before seeing returns, and that’s scary.
First, the good news is that many subscription apps see a large share of LTV on day 1. Why? Because weekly and monthly subscriptions often don’t last long. People cancel. So the revenue you see in the first day or week is often a big chunk of what you’ll ever get from that cohort.
My rule of thumb is to model weekly and monthly subscribers as if you’ll get roughly your yearly price back (or more) from the average subscriber over time. This helps you set realistic ROAS targets.
To speed up signal:
- Push yearly plans harder (especially if you have strong retention)
- Consider lowering your yearly price to pull more people into prepaid
- Hard paywalls or no trial mean faster conversion signal and less guessing. I tested this by removing the trial for paid Meta traffic on one app
Set ROAS targets for D30 or D60, not 6 months. Waiting in the dark for half a year isn’t a strategy.
Adapty also offers a lending solution that gets you your Apple money faster than the usual 6 weeks. Worth looking into if cash flow is a constraint.
AI creatives: volume vs. quality
There’s massive hype around UGC generated by AI right now. “Just pump out 500 variants and let the algorithm find the winner.”
I think this misses the point.
The metric that matters is cost per winner, CPW, not number of videos produced.
If you create 500 AI videos and find 1 winner, that might be worse than creating 30 DIY videos and finding 1 winner, especially if the AI videos cost more to produce or required expensive tools.
That said, creatives made with AI help can absolutely outperform handmade ones. I’ve seen it happen.
The keys:
- Hooking in the first 3 seconds is everything, because if you lose them there, nothing else matters. At a Meta event in Berlin, Meta presented that 46% of conversions happen in the first 2 seconds.
- Clear value props. What does your app actually do for them?
- Realistic UI from inside the app. Show the actual experience, not abstract concepts.
- Subtitles. They help way more than people think. Most people watch without sound.
Don’t scream “APP APP APP” in the first second. Start with a pain, an emotion, a moment they recognize. Then show how your app fits in.
One tool I like is Higgsfield AI. It makes mistakes but creates good hooks. Skip expensive AI studios unless they’re actually improving your cost per winner.
The bottom line
- Creative first. Nothing else matters until your IPM is at least 2.
- ROAS is queen (or king). Cheap installs mean nothing without conversion.
- Trust the algorithm, mostly. But know when to override it (trials, especially).
- Prove winners where you’ll scale. A Tier 2 round is fine for cutting weak versions.
- Move fast. 3 days tells you whether a campaign is alive. D30 tells you whether it pays.
This playbook took me years and a lot of burned budget to figure out. Hope it saves you some pain.