Do I need a Meta Conversions API expert for my app?

I am Samet Durgun, a fractional Head of UA. I run paid UA for subscription apps and mobile games and write up what I find in the accounts I manage. This piece sits under what signal engineering fixes; more about me.

You may not need a Meta Conversions API expert yet. First find out where the event breaks: Meta never received it, it carries the wrong name or value, it is counted more than once, Meta cannot match or use it, or it is not the event your campaign optimizes on. A few test events show which. I write from app accounts: subscription apps and mobile games. If you run a website lead form, the website column in the checks table and one section near the end are for you, and both are my reading of Meta’s documentation. I will not say where to hire. I can tell you what to check and what to ask.

When a client says their Meta events look wrong, the attribution engineer I work with and I start with the billing system, then the mobile measurement partner (MMP), then Events Manager, and look for the first place they disagree. I decide what each event means and check it against the money.

Scope: subscription apps and mobile games on Meta, with events from an SDK, an MMP or a subscription platform, plus web to app campaigns, where the ad opens a web page before the app and purchases reach Meta through the Conversions API. I have not run Meta’s Conversions API Gateway, sent website lead form events through the Conversions API, or worked for a debt relief advertiser. I sell this work under signal engineering, so keep that in mind.

What does a Meta Conversions API expert do?

On an app, “Conversions API expert” covers two jobs, and a third person ships the result.

Who What they check What they hand you
The signal owner (the signal is the events and values Meta receives from you) Which events exist, what each means, which system sends each, and whether counts and values match revenue An event inventory, a written fix request for your developer, a mapping from each event to each platform, and a revenue comparison you can run again
The attribution engineer When the SDKs start, whether each event is logged once, SDK versions and calls, the setup of Apple’s tracking prompt (App Tracking Transparency, ATT), server calls and IDs Findings, recommended fixes and a note of what cannot be fixed
Your developer How the fix fits your code The shipped change

I am the first row. I design the event setup and answer for it, and I do not write SDK or server code. My fix request has four parts: the symptom quoted from Events Manager, a question that lets the developer confirm the cause, one fix, and what that fix gives up.

On the accounts I work on, app purchases often reach Meta through an integration that the subscription platform (the tool that handles in app billing) or the MMP already offers, so the work is settings plus the order in which the SDKs start. For websites I can only quote Meta: it calls its hosted setup an option that “requires no technical expertise, no costs, no ongoing maintenance”, and its Gateway page describes a setup that needs no coding.

I have said before that a full time hire for one narrow skill can go wrong. “Conversions API expert” names one narrow skill too. The wider question is who to hire for your subscription app’s mobile UA.

Where can a Meta event break?

I run five checks in order and stop at the first that fails.

Check What you see On an app, look here On a website, look here (from Meta’s pages) Who fixes it
1. Received Missing, or far below the billing system The sender’s delivery log, then Events Manager What the Pixel fires in the browser, then Events Manager Engineer, developer
2. Named and valued right Right count, wrong name, value or currency The mapping in the MMP or subscription platform, and one raw event What the Pixel fires, which Meta says its Gateway receives Signal owner, developer
3. Counted once Totals above the billing system Every sender, and whether the copies share an event ID The event name and ID on both copies Signal owner, developer
4. Matched or eligible Accepted, but the campaign cannot use it Aggregated Event Measurement (AEM) eligibility, and the fields your app must send Event Match Quality, which Meta offers for web events only Engineer, developer
5. The event the campaign optimizes on All arrives, but the campaign learns from another event The campaign’s optimization event, and the integration chosen for it in Events Manager The same question, my reading Signal owner

My post on Meta AEM eligibility covers checks 1 and 4. Checks 2, 3 and 5 are new here. The event the ad platform optimizes on has to be the one the business counts.

How do I test it before I hire anyone?

For app events, I check one path end to end in four steps before reading any totals, as I did for web to app in my trace of one test click.

  1. Build an inventory: one row per event per dataset (Meta’s container for your events), with source, status, 28 day count and last received time. On one subscription app with a web dataset next to the app dataset, the inventory alone showed three faults: every web purchase carried the same value, two events counted nearly the same thing through different integrations, and one event was marked active with nothing arriving.
  2. Send more than one test event. My check is a paid purchase, an event that should carry no value, and a second purchase from the same user. Each has to land with the value I intended.
  3. Write down the time, then check the sender’s log and Events Manager. For events sent to its API directly, Meta says you “should be able to verify them within 20 minutes”. RevenueCat says events can take up to 24 hours through its integration, so check your sender’s page before you call an event missing. A test code does not keep the event out of your real data: Meta says events sent with test_event_code “are not dropped” and “are used for targeting and ads measurement purposes.”
  4. Ask the developer for the exact event name, including upper and lower case, the source, the value and currency, the counts for the last week and the last 28 days, and whether two events fire at once.

On one iOS app with an MMP, the engineer reviewed SDK start, event logging and the timing of Apple’s tracking prompt, and found why Meta saw too few installs. The developer shipped three fixes: the MMP’s SDK starts when the app opens, each event is logged once, and the tracking prompt shows at app start. Afterwards the same registration event came out almost equal in the MMP and in Meta for the same week, and Meta saw a much larger share of App Store installs. That was one week, so my read was only that the tracking looked fixed.

Sometimes a review finds that every event arrived. On one iOS subscription app, the MMP showed purchases for a campaign and Meta’s campaign numbers showed none, yet the billing system, the MMP and Events Manager showed nearly the same totals. The gap was in how purchases were attributed to the campaign. You can then stop checking delivery and look at attribution.

Who should I hire, and what should I ask?

This is my own list, not an industry rule. I look for someone who:

  • owns the whole event path, from the app SDK and the MMP to the backend and Events Manager, because most Meta signal bugs sit where two parts meet
  • shows how they approach your problem, because I do not think case studies tell you who is good
  • has tracked both web and app, because the two are tracked differently
  • says “this is not in my scope” when it is not
  • treats a signal fix as a hypothesis, judges it on return on ad spend, and promises no result
  • asks first what changed in the account recently

Two of my hiring questions for a mobile game fit here. Ask what they will do themselves and what they need from your engineers. Ask what they do when Meta, the MMP and your backend disagree, and look for a reconciliation method. If a candidate promises to bring attribution back after ATT, ask where the user ID comes from. If you are not sure you need a channel specialist at all, my post on fractional CMOs and channel specialists compares the two.

Before they start, hand over the event names exactly as sent, every source that sends each event, one test event’s request and response, and a dated log of every change. Give view only access first; whoever configures the Conversions API needs admin rights on that one dataset and nothing more.

Should the same event come from two places?

It depends on whether both copies carry the same event name and event ID. Meta deduplicates on the two together, and only for copies received within 48 hours of the first one. Its developer page says the Pixel’s event “must match” the Conversions API’s event_name, so a Lead from the browser and a Contact from the server for the same form are two events to Meta, even with the same ID. Events Received in Events Manager is “the sum from both sources, not deduplicated.”

For a website, Meta wants two senders. Its best practices page says: “Use the Conversions API in addition to the Meta Pixel, and share the same events using both tools.” As I read Meta’s pages, one system then creates the ID for both copies. For app events that feed value optimization, Meta’s advice points the other way: About Partner Integrations for app events says to send an event through one channel only, “either FB , MMP or CAPI”.

On one app, the developer compared the IDs that two senders attached to the same event. They looked nothing alike, and I asked why we kept both senders. Two senders can work. One client of mine runs an MMP and the Conversions API side by side, but it took engineering to get there.

The sources disagree on apps. Meta’s Conversions API for App Events page says it “applies the same deduplication functionality” as for web events. AppsFlyer, on a page about running two MMPs that was last updated about two years ago, says Meta deduplicates installs across MMPs but not in app events. RevenueCat’s Meta Ads page says the Meta SDK and RevenueCat “don’t coordinate a shared event_id”. My read: Meta can remove duplicates on apps, but the common app tools do not send a shared ID, so when two tools send the same event, Meta may count it twice. Meta’s single integration page comes close to saying so: an app event sent through multiple integrations “might be counted multiple times”.

If your own server sends the event, ask what happens when a request fails and is sent again. Meta’s Using the API page says that if it receives “the same event_id and event_name more than once, we keep only the first copy.” So a retry that keeps its ID is safe. As I read it, one with a new ID counts as a second event.

Does a received event mean Meta can use it?

No. Meta’s best practices page says: “You can use only matched events for ads attribution and ad delivery optimization.” Matched, as I use it, means Meta could tie the event to a person’s Meta account.

The same goes for the response your server gets back. Meta’s page on troubleshooting the Conversions API says “If the event payload is valid, a 2xx HTTP response code is returned.” A success code tells you the payload was well formed and says nothing about naming, counting or matching.

On the web side the measure is Event Match Quality, Meta’s score out of 10 for how well events match. Meta recommends “6.0 or higher” and says the score “is currently available only for web events”. On web to app I aim higher. That is my preference, not Meta’s number, and the test is whether return on ad spend improves. An email you already collect with consent, sent hashed as Meta specifies, is a big plus for matching, and without an identifier like that no vendor can work magic. Whether you may send it is a question for your privacy adviser. Even a good score does not get everything back on iOS.

For app events Meta says the score is not available and points you to your Meta representative. There I check AEM eligibility, plus one setting in Events Manager: which integration Meta uses for an app event when several send it. I checked this setting on one iOS campaign that showed zero installs while purchases appeared, and changing it did not solve the case. Meta’s help page on choosing a single integration says events from the chosen one “may be prioritized for accuracy.”

Why can two numbers that agree be bad news?

It depends on what the two numbers count.

The same event, counted by two systems for the same week, should agree. Registrations in the MMP and in Meta are an example, and after a fix that agreement is the first thing I look for.

Meta’s count for its own campaigns should be smaller than your total, meaning everything the billing system saw from every source. When the two match, someone is sending Meta everything, or sending it twice. The exception is an app whose users nearly all come from Meta.

So when a number suddenly looks great, I assume that either things are going very well or there is a mistake, and I check a second source. How Meta’s count ends up too high, and how to check it, is in my post on missing iOS trials.

As I wrote about the billing system, the MMP and Meta, making all three tools agree is the wrong goal. I want a gap I can explain.

What goes wrong after the event arrives?

Values and meaning go wrong. These are patterns from my accounts.

  • On the Google account behind my bid strategy test, a revenue event sent $1 when the real value was $0, which inflated the ROAS Google reported.
  • A value on something that did not happen. On a trial start my rule is to send no value, because nobody has paid yet.
  • In one audit, an event named “trial” carried a purchase value.
  • In one game, an event almost certainly changed its meaning after an app release. In the same game, test purchases were counted in one tool and not in the other, and a missing parameter made a value show as zero.
  • Local currency labeled as dollars, refunds or renewals not sent, and a reporting tool that quietly switched from gross revenue to proceeds. My diagnostic for rising CPI and stalled ROAS has the longer list.

Some of these do not get solved, and I have one of my own. The iOS campaign with zero installs went to the platform’s representative and to the MMP’s support. Settings were changed. Weeks later it was still open.

When do I not need one, and what can a fix break?

You may not need one yet. My own line is “Don’t obsess over signal setup.” A simple app with a simple package is fine as long as the campaign optimizes for value. On one game, the subscription platform already sent purchases to Meta through its own Conversions API integration. When the team asked about sending events from their own backend, I said it did not matter yet.

My own practice is less tidy. My signal engineering page calls signal the first thing I fix on every account, and the basic check always comes first: each event arrives once, with the right value. Beyond that, I say signal work waits for volume, and I have still asked early for a custom event that combines two events, on more than one account.

A broader event, such as a trial plus one more step, only makes sense to me when that step goes together with paying later. When I try one, I change one thing and write down the date.

A fix can also break things.

  • What the campaign learned on bad data can outlast the bug. That is my view of how campaigns behave, and I have not measured it.
  • A fix can hide users you used to see. On one account I wrote a request for the developer to start the MMP’s SDK only after the ATT answer, and named the tradeoff: users who leave before that point are no longer tracked as installs. The tracking review above went the opposite way, so copy neither fix without a review.

What carries over to a lead form on Meta’s Gateway, and what does not?

On apps I treat a signup as a weak signal and ask for events closer to money. On one iOS app, Meta received only the signup, so it could not optimize for anything later. A form lead looks like the same case to me: ask what happens to the lead after the form before you pay anyone. As I read Meta’s description, a setup that only copies what the Pixel sees does not know what happens later in your CRM.

I expect the five checks to apply. Start with what the Pixel fires, because Meta says the Gateway takes its events from the browser. A doubled Events Received total is not proof of duplicates, because it adds both sources before deduplication. If duplicates remain after that, list every tool that writes to the dataset. A test lead is a real lead and a real event, so warn whoever follows up on the leads.

Two points about your category, from Meta’s pages:

Whether these rules apply to you is for your legal adviser; nothing here is legal advice. For this setup, hire someone who has done web lead generation in a regulated category. I have never seen a dataset in core setup. For a lead form behind Meta’s Gateway, I have no first hand work to show.

Sources and scope

I read every page below on 14 November 2026. The 20 minutes, 48 hours, 6.0 and seven days are Meta’s numbers, the 24 hours is RevenueCat’s, and aiming above 6.0 on web to app is my preference. The cases come from a handful of accounts, several from one account, all anonymized.

Questions people ask

Does the Conversions API fix attribution after ATT?

No. For app events the Conversions API fills gaps. It can send events the app cannot see, such as renewals and trial conversions that happen while the app is closed, and it can send their value. Meta still requires the status of Apple's tracking prompt (App Tracking Transparency, ATT) on every app event, so the Conversions API does not bring back attribution for users who declined tracking.

Does Meta remove duplicates when two tools send the same event?

Only when it can match the copies. Meta's main rule, on its deduplication pages, is that both copies carry the same event name and the same event ID, and that the second copy reaches Meta within 48 hours of the first. For a Pixel and server pair Meta also describes a weaker fallback that uses external ID or fbp. On the app tools I work with, the two senders often do not share an ID, so my first question is whether both copies carry the same one.

What Event Match Quality score should I aim for?

Meta recommends 6.0 or higher out of 10 and says the score is currently available only for web events. For app events Meta says the score is not available and points you to your Meta representative. The check I use there is eligibility for Aggregated Event Measurement.

Is a Conversions API expert a marketer or an engineer?

Both, on an app. One person decides what each event means and checks it against revenue. An engineer checks the SDK and server code, and you need one developer who can ship the change. I do the first job and bring in the second, an attribution engineer from my network. For a website that only sends a server copy of its Pixel events, Meta describes setups that need no coding.

How long until a Conversions API fix shows?

It depends on the step. Meta says you can verify a sent event in Events Manager within 20 minutes, and RevenueCat says events sent through its Meta integration can take up to 24 hours to show. After you select an integration for an app event, Meta says the new check for Aggregated Event Measurement eligibility can take up to seven days. What the campaign learned on bad data can outlast the fix, and I cannot give a number for that.