A lower CPI makes a subscription app less profitable when the cheaper installs convert to trials and paying subscribers at a lower rate than the installs they replaced. Cost per paying subscriber is CPI divided by your install-to-paid conversion rate, so a 20% cheaper install is only a saving if install-to-paid falls by less than 20%. Cheap installs usually come from curiosity-led hooks, low-intent audiences, cheaper countries, Android or low-attention placements, and every one of those tends to lower the conversion behind the install. Judge a creative or campaign on cost per subscriber and, where you can, cost per renewed subscriber. CPI is an input, not a result.
This guide covers the break-even rule, a worked example in which the cheaper creative wins every early metric and loses the one that pays the bills, the six places cheap installs come from, and a five-step check you can run on your own account. If you want the definitions of CPI, CPA and CAC first, start with our guide to which subscription app metric actually matters.
Key takeaways
- Cheap installs are not necessarily good installs. A lower CPI only helps if downstream conversion holds.
- The break-even rule is simple: install-to-paid conversion can fall by the same percentage as CPI, and no more.
- The funnel multiplies, so two small drops in trial rate and trial-to-paid can cancel a large CPI saving.
- The gap between a cheap creative and a good one widens the further down the funnel you look, which is why it hides in early reporting.
- Most CPI drops come from a change in who you are reaching, not from the same people getting cheaper.
- Use CPI as a guardrail and diagnostic, never as the target you reward a buyer, an agency or a creative for hitting.
The break-even rule: how much conversion can a cheaper install lose?
Cost per paying subscriber, often called subscriber acquisition cost, is the paid acquisition spend needed to acquire one paying subscriber. For an app with a free trial it breaks down into three numbers.
Cost per subscriber = CPI ÷ (install-to-trial rate × trial-to-paid rate)
A lower CPI reduces cost per subscriber only if the combined conversion rate falls by a smaller percentage than the CPI did.
That sounds generous. A 30% cheaper install can absorb a 30% drop in conversion. The catch is that the funnel multiplies, so modest drops at each stage compound. The table shows how little it takes.
| CPI reduction | Maximum fall in install-to-paid | Wiped out by, for example |
|---|---|---|
| 10% | 10% | Trial rate down 6% and trial-to-paid down 5% (combined about 11%) |
| 20% | 20% | Trial rate down 12% and trial-to-paid down 10% (combined about 21%) |
| 30% | 30% | Trial rate down 20% and trial-to-paid down 13% (combined about 30%) |
| 50% | 50% | Trial rate down 30% and trial-to-paid down 30% (combined about 51%) |
And that is before renewal. If the cheaper subscribers also cancel sooner, a fourth multiplier joins the chain and the bar drops again. A 20% fall in trial rate on its own is easy to miss in a weekly report. Paired with a slightly weaker trial-to-paid rate, it is enough to cancel the largest CPI win most accounts will see all year. Our guide to calculating break-even CPI for a subscription app works the same arithmetic the other way round, from what a subscriber is worth to the most you can pay per install.
Worked example: the cheaper creative that cost 38% more per subscriber
Here is a hypothetical comparison, with illustrative numbers that are not client data. A subscription app spends £6,000 on each of two creative concepts over the same fortnight, in the same country, on the same platform, with the same optimisation event. Creative A leads with the specific problem the app solves and shows the product. Creative B is a broader, more entertaining hook that stops more thumbs and says less about what the app does.
| Metric (hypothetical) | Creative A | Creative B | Better |
|---|---|---|---|
| CPI | £2.40 | £1.60 | B, by 33% |
| Installs | 2,500 | 3,750 | B, by 50% |
| Install-to-trial rate | 9.0% | 5.5% | A |
| Trials | 225 | 206 | A, by 9% |
| Cost per trial | £26.67 | £29.09 | A, by 8% |
| Trial-to-paid rate | 38% | 30% | A |
| Paying subscribers | 86 | 62 | A, by 38% |
| Cost per subscriber | £70 | £97 | A, by 28% |
| First renewal rate | 60% | 50% | A |
| Subscribers after first renewal | 51 | 31 | A, by 66% |
| Cost per renewed subscriber | £117 | £194 | A, by 40% |
Creative B buys 50% more installs for the same money. Its install-to-paid rate is 1.65% against A's 3.42%, so it converts at less than half the rate, and a 33% CPI saving cannot cover a 52% conversion gap. A turns £6,000 into 86 paying subscribers; B turns it into 62. After the first renewal, A has 51 subscribers left and B has 31.
Why the gap hides in early reporting
Look at how the lead changes as you move down the table. B wins CPI by 33%. A wins cost per trial by only 8%, close enough to look like noise in a weekly report. A wins cost per subscriber by 28%, and cost per renewed subscriber by 40%. The difference between a cheap creative and a good one is smallest where the data arrives first and largest where it arrives last.
That is the trap. On day three, B looks like the winner and A looks marginally better at best. Trials resolve within a few days or weeks, depending on trial length, and renewals take a full billing period. A team that scales on CPI or early cost per trial will move budget to B, and the cost of that decision will not be visible until the cohort matures. This is why we read Day-7 ROAS as an early signal, not a verdict, and why our subscription app ROAS benchmarks compare D7 with D30 and D60 before anyone calls a winner.
The further down the funnel you look, the wider the gap between a cheap install and a good one. Decide on the furthest number that has matured by the time you need to act.
Where cheap installs come from
When CPI falls, the first question is not "how do we get more of this?" but "who are we reaching now that we were not before?". Most CPI drops come from a change in the mix of people, places or promises, not from the same people becoming cheaper to reach. These are the six sources we check first.
| Source | Why CPI falls | What happens after the install | How it shows in the data |
|---|---|---|---|
| Curiosity-led hooks | Puzzles, shock openers and entertainment get the tap without selling a reason to pay. | Install-to-trial falls first. People open the app to see what it is, then leave at the first question or the paywall. | Strong thumb-stop and CPI, weak cost per trial on the same concept. |
| Promise mismatch | The ad sells an easier or different outcome than the product delivers. | Trials may hold, but trial-to-paid and early renewal fall when the product does not match the promise. | Concept wins on trials, loses once trials expire. |
| Low-intent audiences under install optimisation | Meta delivers the event you ask for at the lowest cost, and the cheapest installers are not the likeliest payers. | Every downstream rate is lower across the campaign, not just one ad. | Install-optimised campaign beats trial-optimised on CPI, loses on cost per subscriber. |
| Country mix shift | Broad or worldwide targeting moves spend towards markets with cheaper inventory. | Revenue per install falls faster than CPI. RevenueCat puts median D14 RPI about five times higher in North America than in India and South East Asia. | Blended CPI drops while spend share by country changes. |
| Platform mix shift | Android installs are usually cheaper than iOS. | RevenueCat's median D35 download-to-paid is 2.6% on the App Store against 0.9% on Google Play. | Blended CPI drops as the Android share of installs rises. |
| Low-attention placements | Some inventory produces quick, accidental or low-attention taps at a low price. | High install-to-open drop-off and very low trial rates on those placements. | Placement breakdown shows cheap installs concentrated in a few placements. |
Countries and platforms: the CPI drops that are easiest to misread
Country and platform shifts deserve their own warning, because they can lower your blended CPI without a single creative changing. RevenueCat's State of Subscription Apps 2026 reports a median revenue per install at day 14 of $0.38 for North American users against $0.08 for users in India and South East Asia. Revenue per install is total revenue divided by total installs, so it already accounts for conversion. On those medians, an install in the lower-value region would need to cost less than about a fifth of a North American install just to return the same revenue per pound in the first fortnight.
Platforms tell the same story. The same report puts median download-to-paid conversion within 35 days at 2.6% on the App Store and 0.9% on Google Play, roughly 2.9 times higher on iOS. An Android install that is half the price of an iOS install is not a bargain on those numbers. We worked through the full store maths, including fees, in iOS vs Android subscription economics. These are medians across many apps, and your own ratio may be very different, which is exactly why you should measure it rather than assume it.
The optimisation event: when Meta finds you cheap installs
Meta's delivery system looks for people likely to complete the event you optimise for, at the lowest cost it can find. Optimise for installs and it will get very good at finding people who install. Nothing in that instruction asks for people who will pay. That does not make install optimisation wrong. With too few trial or purchase events, it can be the only setting that lets a campaign learn. It does mean an install-optimised campaign has to be judged on downstream cost, not CPI. Our guide to whether to optimise Meta Ads for installs, trials or purchases sets out the signal volumes that make a deeper event workable.
The creative: hooks that sell the tap, not the product
The most common source in our experience is the creative itself. An ad built purely to stop the scroll can earn a very low CPI by attracting people who are curious rather than people who have the problem. The same applies when an ad promises an outcome the onboarding and paywall do not follow through on. We covered that second case in depth in ad-to-paywall message match. A good hook earns attention from the right people. A cheap hook earns attention from anyone.
When a lower CPI genuinely is better
None of this means a lower CPI is suspicious by default. Plenty of CPI improvements are real, and they are worth chasing. A lower CPI is usually good news when:
- The new creative makes the same promise to the same person, only more clearly or more persuasively.
- Your App Store or Play Store page converts better, so more of the same visitors install.
- You have replaced fatigued ads with fresh ones and costs have returned to where they were.
- Country, platform and placement mix are unchanged, and cost per trial has fallen roughly in line with CPI.
If those hold, lower CPI is the most direct lever you have, and our playbook on how to lower CPI on mobile apps covers how to pull it. For context on what a normal install costs in your category, see our mobile app CPI benchmarks.
How to check whether your cheaper CPI is making money
This is the check we run when a concept or campaign suddenly becomes cheaper. It takes about half an hour with access to Meta Ads Manager and your subscription analytics.
- Compare like with like. Break CPI down by country, platform and placement before comparing it with last month. If the mix has moved, compare each segment with its own history, not the blended figure.
- Read cost per trial within 48 hours. Trials start fast. Adapty's State of In-App Subscriptions 2026 reports that 89.4% of trial starts happen on the day of install, so cost per trial by concept is readable almost as quickly as CPI. If CPI fell 30% and cost per trial did not move, the cheaper installs are not starting trials.
- Wait for trial-to-paid before scaling hard. Hold the cheaper concept at a steady budget until its first trial cohort has converted or cancelled, then compare cost per subscriber.
- Check the first renewal. On monthly or weekly plans, compare renewal by acquisition concept or campaign where your data allows it. This is where curiosity-led subscribers usually show themselves.
- Compare revenue per install with CPI. Revenue per install already includes conversion, so its ratio to CPI is the cleanest single comparison between two sources of installs.
One practical limit: on iOS, Apple's privacy rules mean ad-level downstream data from Meta is aggregated and delayed, so concept-level trial and paid comparisons are often clearer on Android, on web-to-app funnels, or through your measurement partner. Our explainer on SKAdNetwork postback delays covers why the iOS numbers arrive late.
Revenue per install matters more than CPI for a subscription app. Two sources of installs can be compared fairly only once conversion is inside the number.
What this means for how you brief and judge creative
If CPI is the number a buyer, an agency or a creative team is rewarded for, it is the number they will improve, and the cheapest way to improve it is usually to reach less valuable people. Set CPI as a guardrail instead: a ceiling that triggers a look, not a target that earns praise. Judge concepts on cost per trial first, cost per subscriber once trials resolve, and early renewal when you have it. Our guide to calculating subscription app CAC covers how to match spend and subscribers to the same cohort so that comparison is fair.
The same logic applies to briefs. A brief that asks for "a cheaper hook" invites curiosity bait. A brief that names the person, the problem and the reason to pay invites a hook that earns attention from people who will convert. In our experience the second kind often has a higher CPI and a lower cost per subscriber, and that is a trade worth making every time.
How we approach it
We build creative around the specific person who pays and the problem they are trying to solve, then judge every concept on trial and subscriber cost rather than install price. Our Oli Help case study is a published example of that approach: ICP-aligned angles for parents of children with ADHD, scaled once they worked, with a 1750% increase in trials and a 394% increase in revenue. You can see how the system fits together on our performance creative agency and Meta Ads service pages, and in how we work with consumer and subscription apps. If your installs are getting cheaper while your subscribers are not, our pricing page shows how we work with apps already spending on Meta. For where paid acquisition sits in the wider funnel, start with our subscription app marketing strategy guide.
Frequently asked questions
Why is my CPI low but my ROAS poor?
Usually because the cheaper installs convert worse after the install. Cost per paying subscriber is CPI divided by install-to-trial rate and trial-to-paid rate, so if those rates fall by more than the CPI did, each subscriber costs more even though each install costs less. The common causes are curiosity-led hooks, a shift towards cheaper countries or Android, low-intent audiences found under install optimisation, and ads that promise something the onboarding and paywall do not deliver.
How much can conversion fall before a cheaper CPI stops being worth it?
By the same percentage the CPI fell. If CPI drops 20%, your combined install-to-paid conversion can fall by up to 20% before cost per subscriber is worse. Because the funnel multiplies, two modest drops add up quickly: a 12% fall in trial rate and a 10% fall in trial-to-paid together cut install-to-paid by about 21%, which wipes out a 20% CPI saving. If the cheaper users also renew less, the bar is lower still.
Is cost per trial a good enough check on install quality?
It is a good first check because it is fast: Adapty reports that 89.4% of trial starts happen on the day of install, so cost per trial is readable within a day or two. It is not enough on its own. Trial-to-paid and early renewal are where low-intent users drop out, and in our worked example the cheaper creative cost only 9% more per trial but 38% more per subscriber.
Should I move budget to cheaper countries to lower CPI?
Only if value per install falls by less than the CPI. RevenueCat's State of Subscription Apps 2026 reports median revenue per install at day 14 of $0.38 for North American users against $0.08 for India and South East Asia, roughly five times higher. A market needs installs at least that much cheaper just to return the same revenue per pound spent in the first two weeks, before differences in renewal.
Should I stop optimising Meta campaigns for installs?
Not automatically. Install optimisation can be the right choice when you have too few trial or purchase events for Meta to learn from. The risk is that Meta will find the cheapest people who install, which is not the same as people who pay, so judge install-optimised campaigns on cost per trial and cost per subscriber rather than CPI, and move the optimisation event deeper once you have the signal volume.
The short version
A subscription app does not sell installs, so it should not buy them on price alone. A cheaper install is a saving only when trial and paid conversion fall by less than the CPI did, and because the funnel multiplies, small drops at each stage add up faster than most dashboards make obvious. When CPI falls, ask who you are now reaching, read cost per trial within two days, hold budget until trial-to-paid has matured, and compare sources on revenue per install. The creative that wins on CPI is often not the one that wins on subscribers.