The CAPI mistakes that silently break your signal

A broken CAPI setup does not throw an error. It feeds Meta duplicate or partial data for months while you assume everything works. These are the most common mistakes and the checks that expose them.

The most dangerous CAPI mistakes are the ones that never throw an error. Your setup appears to work, events come in, the interface shows green checkmarks, and meanwhile Meta is optimizing on duplicate conversions or missing half your data. For B2C companies spending serious budget, that is not a detail: every campaign decision, the algorithm's and yours, is only as good as the signal underneath it. This article covers the mistakes we run into most often and the checks that expose them.

What does CAPI actually do for your account?

The Conversions API sends events directly from your server to Meta, alongside the events the pixel sends from the browser. That matters because the browser route fails more and more often: ad blockers, browser restrictions and declined consent mean a share of your conversions never reaches the pixel. The server route fills those gaps. More complete data means better signal, and better signal means the algorithm can decide more precisely who should see your ads. But that double route also introduces the biggest risk: the same event now arrives twice.

Mistake one: deduplication that does not actually work

Meta deduplicates browser and server events based on event name plus event ID. If your pixel and your server send the same purchase event with the same ID, Meta counts it once. If the ID differs, is missing on one side, or the two routes fire at slightly different moments with different parameters, Meta counts two purchases where there was one. The result is an account that looks better than it is: ROAS in the interface goes up while your bank account keeps telling the same story.

This is the mistake that stays hidden longest, precisely because it improves your numbers. Nobody raises the alarm when results go up. The check is simple: compare the number of purchase events in Events Manager with the number of orders in your backend over the same period. If those consistently diverge, you have a deduplication problem or a gap problem, and you want to know about both.

Mistake two: low event match quality

A server event is only valuable if Meta can tie it to a person. It does that using customer parameters: email address, phone number, name, location and click identifiers. The more correct parameters you send along, the higher your event match quality and the better the odds the conversion gets attributed to the right user. Setups that only send an email address, or deliver parameters in the wrong format, throw away a large part of CAPI's value without anything visibly breaking.

Match quality is not a setting but an outcome. You improve it by sending more parameters at the points where you already have them, such as checkout, and by verifying they arrive correctly hashed and formatted. Events Manager shows a score per event along with the parameters that are missing. Reviewing that score once in a while is the difference between a CAPI setup that adds signal and one that only adds server costs.

Mistake three: only sending purchase events

Many setups only send the purchase event through the server, because that is the event campaigns optimize on. But the algorithm also learns from the steps before it: content views, add to carts and checkouts reveal who is on their way to a purchase. If those events are missing on the server side, they disappear for exactly the users who block the browser route, and that group keeps growing. Send your whole funnel through both routes, with deduplication in order, and the algorithm sees the complete picture again.

A tracking bug that improves your numbers is the most expensive bug there is.

Which checks catch these mistakes?

You do not need to be a developer to find a broken setup. Run these checks every quarter, or immediately after any change to your store or tracking:

• Compare purchase events in Events Manager with orders in your backend over the same period: more events than orders points to double counting, fewer points to gaps.
• Check per event whether both the browser and the server route fire and whether Meta shows them as deduplicated.
• Review the event match quality of your key events and which parameters are missing.
• Test a purchase with an ad blocker enabled: does it still arrive through the server?
• Repeat the checks after every theme update, app switch or checkout change, because those are the moments setups silently break.

Conclusion

CAPI is not a project you deliver once, it is infrastructure you inspect periodically. Deduplication, match quality and full-funnel events together determine whether Meta sees your reality or a distorted version of it. And because every media buying decision rests on that signal, tracking belongs to the core of good account management, not to the afterthoughts. That is why, on every account we manage, we get the signal right before we touch anything else.

Not sure whether your setup does what it promises? Book a call and we will gladly look at your events and your data with you.