To audit your mobile app marketing agency, gather the evidence before you talk to them, then score ten areas: ownership, transparency, attribution, reporting, unit economics, strategy, creative, account structure, speed and incrementality. The evidence is mostly things you can pull yourself: who owns the ad account and app in your business settings, 90 days of activity history in Ads Manager, the last three monthly reports, a list of ads launched with their spend, and your own subscription revenue by cohort. The single most revealing step is putting the agency's reported revenue next to your own revenue data for the same cohorts. If that comparison has never been made, it is the audit's first finding.
Most guides to auditing an agency are written for businesses where a click becomes a sale the same day, and they mostly check invoices, reports and meeting notes. A subscription app needs more than that, because the revenue from this month's spend is still arriving six weeks from now, and because the numbers in an ad platform report are not the numbers that reach your bank. This guide is for apps that already have an agency and want to know whether it is working. If you are still choosing one, start with our questions to ask a Meta Ads agency instead.
Key takeaways
- Collect the evidence first. Most of it lives in accounts you should already own.
- Score ten areas from 0 to 2. Ownership, transparency and attribution are gates: a zero on any of them needs fixing whatever the total.
- Reconcile reported revenue with your own revenue data. Reported ROAS and money in the bank can differ by more than a third.
- Judge only cohorts old enough to have matured, roughly six weeks or more for a subscription app.
- Audit your own side too. Product, paywall and price changes move results as much as the agency does.
When to audit, and when it is too early
The first 90 days are a different exercise. In that period you are checking whether the setup, the research and the first tests are happening, and our 30/60/90 day plan for a new app marketing agency has a checkpoint scorecard for exactly that. This audit is for after the first quarter, when there is enough history to judge outcomes rather than activity.
After that, run it about twice a year, and also at four moments when the cost of not knowing is highest: before a contract renewal, before a large budget increase, when a new head of growth or founder takes over the relationship, and when results have been weak for more than a month. A single bad week is not a trigger. At any given moment, the newest cohorts in a subscription app have not finished converting, so a week of weak numbers is mostly noise.
The agency audit scorecard
Score each area 0, 1 or 2. A 2 means the evidence clearly shows the good version, a 0 means it shows the bad version or the evidence does not exist, and a 1 is everything in between. The evidence column is what to ask for or pull yourself.
| # | Area | Evidence | Scores 2 | Scores 0 |
|---|---|---|---|---|
| 1 | Ownership | Business settings for your portfolio; admin list on the measurement partner; where the raw creative files live. | Your business owns the ad account, app, dataset and files. The agency has partner access. | The ad account or app sits in the agency's portfolio, or the raw files are theirs. |
| 2 | Transparency | Your own access level; how media and fees are invoiced; activity history. | You have admin access, media is billed to your payment method and fees are invoiced separately. | View-only or no access, or media and fees arrive as one lump sum. |
| 3 | Attribution | Install, trial and purchase counts in Events Manager against your own data for the same days. | Each event arrives once, and the gap to your own data is known and explained. | Nobody has ever compared them. |
| 4 | Reporting | The last three monthly reports. | Cohort revenue from your revenue data, the window stated, store fees accounted for. | Cost per install, click-through rate and platform ROAS only. |
| 5 | Unit economics | Cost per subscriber by monthly cohort; cost at the last budget increase; payback against your target. | Known, inside target, and the cost of the extra spend is tracked separately from the average. | No cost per subscriber, or only an account-wide average. |
| 6 | Strategy | The current written plan and what the agency believes your main constraint is. | A named constraint and a plan that has changed as evidence came in. | “Testing everything” with no stated reason. |
| 7 | Creative | Ads launched each month, tagged by concept, with spend. | A steady flow of genuinely different concepts, some of which become the ads carrying spend. | Variations of one ad, or spend still sitting on ads that are many months old. |
| 8 | Account structure | 90 days of activity history. | Structural changes are rare, dated and explained. | Frequent rebuilds and ad sets that never leave the learning phase. |
| 9 | Speed | The test log, with dates. | A finding becomes a brief within days and a live test within a couple of weeks. | No log, or findings that take months to act on. |
| 10 | Incrementality | Any lift test, holdout or geo test, or the plan for one. | Tests run where volume allows, and the result changed a decision. | Never raised, even though the account qualifies. |
Step 1: collect the evidence before the conversation
Ask for the evidence in writing, with a few days' notice, and gather your own side at the same time. Doing this before the review meeting changes the meeting. Instead of a presentation you react to, you arrive with questions about specific numbers. The evidence pack for a Meta-led app account is short:
- A screenshot or export of your business settings showing who owns the ad account, the app and the dataset, and who has access.
- The last three monthly reports, unedited.
- An export of ads launched in the last six months with launch date, concept name and spend.
- The test log: what was tested, why, when, and what was decided.
- Your own subscription revenue by monthly install cohort, from RevenueCat or your backend, for the same period.
- The invoices for the same months, so fees and media can be separated.
How the agency responds to the request is itself evidence. An agency with a test log and cohort reporting can send this in a day. One that needs two weeks to assemble it probably does not use it.
Step 2: check ownership and access first
Ownership is first because it decides what you can do with the rest of the audit's findings. If the ad account or the app sits in the agency's business portfolio, a poor audit leaves you with a choice between staying and losing your history. Our guide to switching mobile app marketing agencies has the full table of the eight assets to check, how to check each one, and which of them cannot be moved later. Run through it now, while the relationship is working, because asking for ownership during a dispute is far harder.
Access is the transparency half of the same check. You should have admin access to your own ad account, media should be billed to your payment method, and agency fees should be invoiced separately from media. If media and fees arrive as one number, you cannot tell what share of the money reached Meta.
Step 3: reconcile their numbers with yours
This is the step that most often changes the verdict. An agency report built from Meta's numbers and the revenue in your subscription data are measuring different things, and the gap between them is not a sign of dishonesty. It is a sign of whether anyone is looking.
Take a hypothetical month. These are illustrative numbers, not client data and not a benchmark. The agency spends £20,000 and the report shows £26,000 of purchase value, a ROAS of 1.3. Assume the app sends Meta the gross subscription price including VAT, and all the buyers are UK iOS users in their first subscription year.
| Line | Value | Note |
|---|---|---|
| Purchase value reported to Meta | £26,000 | Hypothetical. Assumes the app sends the gross price including VAT. |
| Less UK VAT (divide by 1.2) | £21,667 | Prices on the App Store include VAT for UK buyers. |
| Less Apple's first-year commission (keep 70%) | £15,167 | Standard rate, first year of a subscription. |
| Return on £20,000 of media | 1.30 reported, 0.76 in proceeds | Before any conversions that arrive after the attribution window. |
The 70% comes from Apple's subscription terms, which pay developers in the UK and US 70% of the price in a subscription's first year and 85% after a year of paid service, with 85% throughout for members of its Small Business Program. Our guide to break-even CPI walks through the same netting for other stores and plan types.
A reported ROAS of 1.3 can be 0.76 in proceeds. If an agency's report has never been reconciled with your own revenue data, you do not yet know whether the account is profitable, and neither does the agency.
Your revenue tool needs the same care. RevenueCat's revenue chart documentation explains that its standard revenue figure is net of refunds but before taxes and store commission, and that its "Proceeds" measure removes both. Compare like with like.
The gap runs the other way too. A platform only credits conversions inside its attribution window, and a subscription app keeps converting long after a 7-day click window closes. RevenueCat's State of Subscription Apps 2026 finds that 19.2% of the paid conversions that eventually happen arrive in week six or later. Which effect is larger depends on your app, which is why the answer is a cohort read from your own revenue data rather than a correction factor applied to the platform's number.
Only judge cohorts that have matured
Because of that tail, the last six weeks or so of cohorts are unfinished at any point you audit. On iOS there is a second clock: Apple's attribution postbacks cover conversion windows up to day 35 and can arrive several days later, so platform numbers for recent iOS installs are still filling in. Score unit economics on cohorts that are at least six weeks old, and compare each one with the same cohort age a quarter or a year earlier.
Step 4: look at the account yourself
Three checks in Ads Manager take under an hour and do not depend on the agency's account of events.
Activity history: how often is the account rebuilt?
Meta's guide to viewing activity history in Ads Manager says Ads Manager "tracks all changes made to your ad campaigns, ad sets and ads", including who made each change and when, filterable by budget, targeting, bid, schedule and run status. Read the last 90 days. You are looking for two things. First, whether the changes match what the reports told you. Second, how often the structure changes. New ad sets and significant edits send delivery back into the learning phase, so an account restructured every few weeks can look busy while never stabilising long enough to learn.
Creative: is new work finding spend?
From the ads export, work out what share of last month's spend went to ads launched in the previous 60 days, and how many of those were genuinely different concepts rather than new colours, voices or edits of one ad. There is no published benchmark for the right share, and it varies with spend and audience size. The trend is what matters. If spend has sat on the same handful of ads for months while new launches get nothing, either the new work is weak or it is not being tested properly, and both are agency problems. Our explainers on creative fatigue and creative diversity versus volume cover how to tell real concept breadth from variation, and the creative refresh calculator gives a rough expectation for how much new work your spend needs.
The test log: can anyone explain the last three tests?
Pick three tests from the log and ask what each one was for, what it found and what changed because of it. This scores strategy and speed at the same time. A good agency can answer for the losers as readily as the winners, because the losers are where most of the learning is.
Step 5: audit the economics at the margin
Average cost per subscriber hides the number that matters when you are deciding whether to spend more: what the last increase in budget bought. If spend rose from £15,000 to £25,000 a month, the question is what the extra £10,000 produced, measured on matured cohorts. An agency that tracks only the account-wide average will show a blended figure that looks healthier than the extra spend really was.
Then compare cost per subscriber with the ceiling your economics allow. If you have not set one, our guides to setting a target CAC and CAC payback show how to work one out. Without a target, no agency can be judged on economics, and that gap is on your side of the table.
Step 6: test incrementality, if you have the volume
Everything above measures what the agency's spend was credited with. Incrementality measures what it caused. Meta's Conversion Lift does this by splitting your audience at random into a test group that can see the ads and a control group that cannot, then comparing conversions. Meta says there is no additional cost for the test, and gives a guide to who can run one: an ad account with a campaign that started in the past year with "a spend of USD 5,000 or more" and "a minimum of 500 conversions" under a 1-day click, 7-day click or 1-day view attribution setting, plus a signal quality requirement for how your events are sent.
For many smaller app accounts that bar is out of reach on a purchase event, and that is fine. Below it, a lift test is unlikely to give a usable answer, and cohort reporting from your own revenue data is the more useful check. Score incrementality as not applicable in that case rather than marking the agency down. Above it, an agency that has never raised a lift test or holdout deserves a direct question. We describe the three levels of measurement, from basic tracking to cohort reporting to causal testing, in our comparison of user acquisition agencies.
How to read the score
Add up the ten scores for a total out of 20. If incrementality is not applicable, score out of 18 and scale.
- Any zero on ownership, transparency or attribution: fix it now, whatever the total. These are the foundations every other score rests on, and all three are quick to put right with a willing agency.
- 16 or more: the relationship is working. Agree the two lowest areas as the focus for the next six months.
- 10 to 15: have a reset conversation. Write down the gaps, the actions and the dates, and re-score in three months.
- Below 10: a reset is still worth one try if the gaps are new, but plan for a switch, and secure ownership before you raise it.
An agency audit is not a verdict on the last month's numbers. It is a check on whether the foundations, the reporting and the process are good enough that the next six months can be judged fairly.
Audit your own side as well
Results in an app account move with the product as much as with the ads. Before blaming the agency for a fall, check whether the paywall, price, trial length or onboarding changed in the period, whether creative approvals took weeks, whether budget was cut mid-test, and whether the agency was asked to fix something it was never hired to do, such as App Store listings or retention emails.
The reverse is true of good results. Our Steps & Beasts case study publishes a 1112% revenue increase and 118% more active subscriptions, and credits an onboarding overhaul alongside heavy creative testing. A fair audit gives the product its share of a result, in either direction. If the honest finding is that the funnel after install is the constraint, a new agency will inherit the same problem.
What we would expect to be audited on
We run paid acquisition and performance creative for consumer and subscription apps, so this is the scorecard we would expect a client to use on us. Our pricing page sets out the terms: monthly reporting is included in every tier, and after an initial three-month commitment the engagement is rolling monthly with 30 days' notice, so a poor audit should never leave a client stuck. Our Meta Ads service puts creative strategy, production and campaign management in one team, which covers areas 6 to 9 of the scorecard. We do not offer app store optimisation or lifecycle marketing, so we would not expect to be scored on either.
If your own audit says the bottleneck is creative or Meta execution, see how we work with consumer and subscription apps, read our case studies with the same scepticism this guide recommends, and apply to work with us. For where paid acquisition sits in the wider funnel, start with our subscription app marketing strategy guide, and for the full buyer's view of the category, our guide to choosing a mobile app marketing agency.
Frequently asked questions
How do you audit a marketing agency's performance?
Collect the evidence before you talk to the agency: ownership and access settings, the last three monthly reports, 90 days of Ads Manager activity history, a list of ads launched with their spend, and your own subscription revenue by cohort. Then score ten areas from 0 to 2: ownership, transparency, attribution, reporting, unit economics, strategy, creative, account structure, speed and incrementality. Reconcile the agency's reported revenue with your own revenue data, judge only cohorts old enough to have matured, and treat any zero on ownership, transparency or attribution as something to fix regardless of the total.
How often should you audit your marketing agency?
Once after the first 90 days, then roughly twice a year, and additionally before a contract renewal, before a large budget increase, or after results have been weak for more than a month. Auditing more often than that mostly measures noise, and for a subscription app the most recent six weeks of cohorts are still converting at any given moment.
Why does my agency's ROAS not match my revenue?
Usually because the two numbers measure different things. Ad platforms report the value your app sends them, often the gross price including VAT, credited within an attribution window. Your bank receives proceeds after VAT and the store's commission. In the hypothetical example in this guide, a reported ROAS of 1.3 becomes about 0.76 in proceeds. The reported figure can also understate, because conversions that arrive after the attribution window are not credited. The fix is a cohort read from your own revenue data, not a correction factor.
Can I run a Meta Conversion Lift test to audit my agency?
If your account is large enough. Meta's guidance says that, as a guide, you need a campaign that started in the past year with at least USD 5,000 of spend and at least 500 conversions under a 1-day click, 7-day click or 1-day view attribution setting, plus a signal quality requirement. Meta says there is no additional cost for the test. Below that size, a lift test is unlikely to produce a usable answer and cohort reporting from your own revenue data is the more useful check.
What are the signs it is time to replace your agency rather than reset?
A zero on ownership or transparency that the agency will not fix, reporting that cannot be reconciled with your revenue data after you have asked, or a low score across strategy, creative and speed that has not moved after a written reset with agreed actions and dates. If the gaps are in your own product, paywall or tracking, a new agency will inherit them, so check that first.
The short version
Gather the evidence before the meeting: ownership settings, activity history, three reports, the ads export, the test log, the invoices and your own revenue by cohort. Score ten areas from 0 to 2, and fix any zero on ownership, transparency or attribution first. Reconcile the agency's reported revenue with your own proceeds, judge only cohorts at least six weeks old, look at marginal rather than average cost per subscriber, and run a lift test only if the account is big enough for one. Then check what changed on your side before deciding whether the agency, the product or both need to change.