Meta, your mobile measurement partner and RevenueCat show different ROAS because they are not measuring the same thing. Meta counts only the users it credits to its own ads, inside an attribution window, at whatever value your app sends it. Your MMP referees between every network that claims the same user. RevenueCat counts the money every customer actually paid, with no window at all. Add different dates, time zones, revenue bases and refund handling, and three honest systems can report three different numbers for the same month. In the hypothetical example below, one month's Meta spend reads 1.20 in Meta, 1.02 in the MMP, 1.33 in RevenueCat at day 60 and 0.80 once tax and commission are taken out. None of them is wrong.
This guide lists every source of the gap, walks through a reconciliation step by step, and sets out which figure to use for which decision. It sits alongside our guides to cohort ROAS versus blended ROAS, which covers the choice between a cohort and a calendar view, and ROAS at D7, D30 and D60, which covers how old a cohort should be before you judge it. This one is about attribution: who gets credit, and for what.
Key takeaways
- Each system answers a different question. Meta ranks Meta ads, the MMP referees between channels, and RevenueCat records what users paid.
- Meta tends to read high on overlap, gross values, refunds and duplicate events, and low on anything that arrives after its attribution window.
- For a subscription app the window matters most, because renewals and late first payments rarely land inside it.
- Do not chase an exact match. Track the ratio between systems and investigate when it moves.
- Make budget decisions on a cohort read from your subscription data, net of tax and store commission.
What each system is actually measuring
Start with what each tool was built to do, because most of the disagreement follows from it. RevenueCat's own documentation is blunt about its role: "RevenueCat isn't an attribution network and can't determine which ad drove an install or conversion," as its AppsFlyer integration guide puts it. It knows what people paid. It only knows where they came from if another system tells it.
| System | Users counted | Window | Revenue value | Question it answers |
|---|---|---|---|---|
| Meta Ads Manager | Only the users Meta attributes to its own ads, under its own rules | An attribution window after a click or view; 7-day click and 1-day view is the common setting | Whatever value your app, SDK or server integration sends with each event | Which of my Meta ads and concepts is doing better? |
| Mobile measurement partner (MMP) | Users it assigns to one source after weighing every network's claims | The lookback windows configured in the MMP, which may or may not match Meta's | The revenue events it receives from the app or from your subscription platform | Which channel gets credit when several claim the same user? |
| RevenueCat | Every customer, paid or organic, unless you filter to an attribution attribute | No attribution window. Revenue keeps accruing for as long as you follow the cohort | Store transactions, with a choice of gross, net of tax, or proceeds after commission | How much money did these users actually pay, and when? |
Meta is a participant marking its own homework, which is not a criticism. Every ad platform does it, and its numbers are still the best available guide to which of its ads work. The MMP exists because several networks can claim the same install, and someone has to decide. RevenueCat sits on the store transactions and is the closest of the three to money in the bank.
The ten reasons the numbers disagree
Most gaps come from a short list. The table shows each cause and which way it usually pushes the figures. Several of them work in opposite directions, which is why two systems can occasionally agree by accident and then drift apart without anything being broken.
| Cause | What happens | Usual effect |
|---|---|---|
| Who counts | RevenueCat sees every customer. Meta and the MMP only count users they attribute to Meta. | RevenueCat higher, unless filtered to Meta-attributed users |
| Competing claims | Meta credits itself for users who clicked or viewed its ads in the window. An MMP picks one winner across networks, so some Meta claims go to another source or to organic. | Meta higher than the MMP |
| Attribution window | A renewal on day 37 or a first payment in week 6 falls outside a 7-day click window, so Meta never credits it. | Meta lower than a mature RevenueCat cohort |
| iOS privacy rules | Apple's postbacks carry no user-level data, arrive 24 to 48 hours after the conversion at the earliest, and can carry less detail for smaller campaigns. | Recent iOS days under-reported in Meta |
| Date basis | RevenueCat cohorts by the date a customer was first seen. Its revenue chart books money on transaction date. Meta dates results by its own reporting convention. | Revenue lands in different weeks |
| Time zone | RevenueCat charts are in UTC. Ads Manager uses the time zone set on the ad account. | Edge-of-day revenue shifts between days |
| Revenue basis | The value sent to Meta can be gross, net of tax, or after store commission, depending on setup. | Meta higher if it receives gross values |
| Refunds | RevenueCat deducts refunds. Its Meta integration does not send negative revenue. | Meta higher by roughly the refund rate |
| Duplicate events | If the Meta SDK logs purchases and a server integration sends the same purchase, Meta can count it twice. | Meta higher, sometimes by a lot |
| Event mapping | RevenueCat sends trial conversions, first purchases and renewals to Meta as Subscribe by default, so a Subscribe count is not a count of new subscribers. | Meta's subscription count looks higher than new payers |
Attribution: who gets the credit
Meta credits a conversion to its ad if it happened within the attribution window after someone clicked or viewed that ad. Seven-day click and one-day view is the common setting. If the same person also tapped a TikTok ad, or would have found the app in the store anyway, Meta still counts them. An MMP looks at all the claims and gives the install to one source. The difference between Meta's total and the MMP's Meta total is mostly overlap and view-through credit.
Apple's own system works the same way at the device level. Its AdAttributionKit documentation says only one impression can win a conversion, that click-through ads always take precedence over view-through ads, and that the most recent tap wins. If you add up each network's self-reported installs, you will usually get more than you actually received.
The window: why subscription revenue goes missing
This is the gap that matters most for subscription apps and the one that works against Meta. A monthly renewal on day 37, or a first payment from someone who started paying in week six, falls outside a seven-day click window, so Meta does not credit it to the ad that brought the user in. RevenueCat's Meta integration can send trial conversions and renewals server to server even when the app is closed, and those events do count toward Meta's revenue figure. Whether they are credited to a campaign is still Meta's decision, under its own rules. RevenueCat's documentation makes that point directly: a successful delivery means Meta accepted the event, not that Meta attributed it.
The practical result is that Meta's ROAS for a subscription app behaves like an early-window reading, whatever date range you pick in Ads Manager. Our guide to choosing between installs, trials and purchases explains why this also affects which event Meta can optimise for.
iOS: delayed, aggregated and sometimes thin
On iOS, attribution for users who have not allowed tracking runs through Apple's privacy frameworks, SKAdNetwork and AdAttributionKit, which Apple documents as working together. Apple's guide to receiving postbacks says the minimum time between a conversion and the ad network receiving the postback is 24 to 48 hours, and its overview says the signed postback does not include user or device data. Later conversion windows arrive later still, which our guide to SKAdNetwork postback delays sets out in full. Meta also uses its own Aggregated Event Measurement for iOS app campaigns.
Two consequences follow. Yesterday's iOS figure in Ads Manager is always the least complete, so never judge iOS on the last two or three days. And because the postback describes a conversion value rather than a named user's transactions, Meta's iOS revenue is an assembled estimate, not a ledger you can match line by line. We would be wary of anyone who claims to reconcile it exactly. Meta documents how its iOS measurement works in its Business Help Centre, and its settings change, so check the current page for your campaign type before relying on any third-party summary, including this one.
Dates and time zones
RevenueCat's charts documentation says all charts are displayed in UTC, and its conversion charts group customers by the date they were first seen. Its revenue chart books money on the date of the transaction. Ads Manager reports in the time zone set on your ad account, so check what yours is set to. A UK account looking at "October" in all three tools can be looking at three slightly different sets of users and payments. Over a month the effect is small. Over a single day, or a week with a big launch on a Sunday night, it is not.
Revenue basis, refunds and duplicates
These three are setup issues, and they are the ones that make Meta look better than it is. RevenueCat lets you choose whether sales are sent to Meta as gross revenue, before store commission, or after commission and estimated taxes, and its documentation advises using the same reporting mode when comparing Meta's ROAS with RevenueCat's revenue. Its integration comparison shows that the Meta integration does not support negative revenue, so refunds never come off Meta's total, while RevenueCat's own revenue figure is net of refunds.
Duplicates are the expensive one. RevenueCat warns that the Meta SDK and RevenueCat do not share an event ID, so if the SDK is still logging in-app purchases automatically while RevenueCat sends the same purchases from the server, Meta can count revenue twice. If your Meta ROAS jumped the week someone connected a new integration, check this first.
Finally, check what is being counted. By default RevenueCat maps trial conversions, initial purchases and renewals all to Meta's Subscribe event, and trial starts to StartTrial. A "Subscriptions" column in Ads Manager can therefore include renewals, and is not a count of new paying users.
A worked reconciliation: one month, four numbers
The figures below are hypothetical, chosen to show the mechanics. They are not client data and not a benchmark. Assume a subscription app spends £20,000 on Meta in a month, sells a monthly plan to UK users on both stores, sends gross values including VAT to Meta, and has no duplicate events. We start from Meta's figure and walk to cash.
| Step | Value | ROAS on £20,000 | Note |
|---|---|---|---|
| Meta Ads Manager purchase value | £24,000 | 1.20 | 7-day click and 1-day view, gross values including VAT |
| Less view-through and overlap claims the MMP gives to another source or organic | −£3,600 | Hypothetical 15% of Meta's claimed value | |
| MMP: revenue from Meta-attributed users, read at day 7 | £20,400 | 1.02 | One referee across all networks |
| Add revenue from the same users after Meta's window, to day 60 | +£6,100 | Late first payments and first renewals, read in RevenueCat | |
| RevenueCat: same users, day 60, gross | £26,500 | 1.33 | Includes VAT and store commission |
| Less refunds at 3% | £25,705 | 1.29 | Meta never receives these as negative revenue |
| Less 20% UK VAT (divide by 1.2) | £21,421 | 1.07 | Store prices to UK buyers include VAT |
| Less store commission at a blended 25% | £16,066 | 0.80 | Hypothetical mix of App Store at 30% and Google Play subscriptions at 15% |
Read the table in two halves. The first half is attribution: Meta claims more users than the MMP gives it, but credits them for less time than RevenueCat follows them. In this example those two effects nearly cancel, which is exactly why a single month's agreement between Meta and RevenueCat proves nothing. The second half is accounting: refunds, VAT and commission take a gross 1.33 down to 0.80 in money the app actually keeps. Our agency audit guide walks through that netting step on its own, and our break-even CPI guide covers the store rates in more detail.
There is no single true ROAS. There are three systems answering three different questions, and the expensive mistake is using one system's answer for another system's question.
Which ROAS to use for which decision
The useful question is not which number is right. It is which number fits the decision in front of you.
| Decision | Use | Why |
|---|---|---|
| Which ad or concept to scale or pause | Meta's own figure | The biases are broadly shared across ads in the same account, so the ranking is usable even when the level is not. |
| How to split budget between Meta and other channels | The MMP | One referee applying one set of rules to every network. Summing each network's self-reported results double counts users. |
| Whether Meta spend is paying back, and how much more to spend | A RevenueCat cohort read, filtered to Meta-attributed users and net of tax and commission | It is the money, followed for long enough to include renewals. |
| Whether Meta caused the revenue at all | A lift test | Every attribution system assigns credit. None of them measures what would have happened without the ads. |
| Cash planning and tax | Store payout reports | RevenueCat itself says to use the payment processor's data for tax reporting. |
One caveat on the first row. Meta's figure ranks ads fairly only when the ads attract similar users. A concept that pulls in people who choose an annual plan will look worse in a seven-day window than a concept that sells monthly plans, even if it earns more by day 60. When concepts sell different plans, check the ranking against a cohort read before scaling the winner.
On the lift test row: Meta's Conversion Lift is a controlled test rather than an attribution model. Meta's eligibility guidance suggests, as a guide, a campaign spending at least USD 5,000 with 500 or more conversions, so it is realistic for many apps already at scale and out of reach for the smallest.
How to reconcile the three each month
A monthly check takes about half an hour once it is set up. The aim is not to make the numbers match. It is to know why they differ and to notice when the reason changes.
- Fix the basics first. Same date range, a note of the time zone each tool uses, and the revenue basis your integration sends to Meta written down somewhere the whole team can see.
- Rule out duplicates. Confirm only one source sends each purchase event to Meta. If both the SDK and a server integration send it, decide which one stays.
- Compare like with like. Compare Meta's trials with RevenueCat trials from users attributed to Meta, not with every trial. RevenueCat's own guidance says not to compare all its trials directly with Meta-attributed conversions.
- Get attribution into RevenueCat. Pass the MMP's final source and campaign into RevenueCat's reserved attributes such as $mediaSource and $campaign. Its documentation warns these can only be set once per customer, so never write a placeholder like "Organic" while waiting.
- Track the ratio, not the level. Divide Meta-reported revenue by RevenueCat revenue for Meta-attributed users at a fixed cohort age. In our example that is £24,000 over £20,400 at day 7, about 1.18. If it sits near that every month, Meta's figure is a usable daily proxy. If it jumps, something changed.
- Leave iOS recent days alone. Exclude the last two to three days of iOS data from any comparison, because Apple's postbacks have not all arrived.
When the ratio moves, the usual suspects in order are: a tracking change such as a new SDK or integration, a shift in the mix of plans or platforms, a change in attribution settings, and a real change in how many users Meta is genuinely driving. Only the last one is a media buying question.
How we report it
Our Meta Ads management reports cohort-level ROAS rather than the platform's figure, and includes MMP and SKAdNetwork attribution setup and QA in scope, because a creative test is only as good as the measurement that judges it. We use Meta's numbers every day to compare ads, and cohort numbers every month to decide spend. If you are not sure which of your three dashboards to believe, that is usually where the first conversation starts.
For the wider picture of how measurement fits into growing a subscription app, see our subscription app marketing strategy guide. If your app already spends at least our £3,000 monthly minimum on Meta and the numbers will not line up, you can apply to work with us.
Frequently asked questions
Why is my Meta ROAS higher than my RevenueCat ROAS?
Usually because Meta credits itself for some users another source would claim, including view-through conversions, and because the value your app or integration sends Meta is often the gross price including VAT and store commission. Duplicate purchase events, sent once by the Meta SDK and again by a server integration, can also inflate it. Check those three before anything else.
Why is my Meta ROAS lower than my RevenueCat ROAS?
Meta only credits conversions inside its attribution window, so renewals and late first payments for a subscription app mostly fall outside it. RevenueCat also counts every customer, including organic users and other channels, unless you filter to users attributed to Meta. On iOS, Apple's privacy-preserving postbacks arrive late and with limited detail, which can leave recent days under-reported in Meta.
Which ROAS should be the source of truth for a subscription app?
No single figure answers every question. Use Meta's own figure to compare ads and concepts inside Meta, the MMP to compare channels against one referee, and a cohort read from your subscription data, net of tax and store commission, for budget and payback decisions. When you need to know whether Meta caused the revenue at all, use a lift test.
Does RevenueCat send refunds to Meta?
RevenueCat's own integration table says its Meta Ads integration does not support negative revenue, so a refund does not reduce the revenue Meta has already recorded. RevenueCat's charts do deduct refunds. That is one reason RevenueCat revenue can sit below Meta's reported value for the same users.
Should Meta and RevenueCat ever match exactly?
No. They use different user scopes, attribution rules, date conventions, time zones and revenue bases, so an exact match would be a coincidence. What should stay steady is the ratio between them. If Meta-reported revenue divided by RevenueCat cohort revenue moves sharply from one month to the next, something has changed in tracking, attribution or the mix of users, and it is worth investigating.
The short version
Meta, your MMP and RevenueCat disagree because they count different users over different windows at different revenue bases. Meta reads high on overlap, gross values, refunds and duplicates, and low on anything after its attribution window, which for a subscription app means most renewals. Use Meta to rank Meta ads, the MMP to referee channels, and a net cohort read from your subscription data to decide spend. Stop trying to make them match. Watch the ratio between them, and treat a sudden change in that ratio as the signal worth investigating.