Five tracking mistakes that make your ad reporting lie to you
Broken measurement is worse than no measurement: it produces confident decisions in the wrong direction. The five failures we find most often, and how to check for each in an afternoon.
There is a particular kind of account that looks like it is working. The dashboards are green, the conversion count is healthy, the cost per acquisition sits comfortably under target. And the business has not noticed any more revenue.
Almost always, the measurement is wrong. Not missing, which would be obvious, but wrong in a direction that flatters. Below are the five failures that cause this most often. Each is checkable in an afternoon and none requires a developer to diagnose.
1. The thank-you page counts more than once
A conversion that fires on a confirmation page will fire again if the visitor refreshes, hits back and forward, or bookmarks the page and returns. On low-volume accounts this can inflate conversions by a fifth without anyone noticing, because the shape of the trend still looks right.
How to check: compare your platform conversion count against actual records for the same period. Leads in the CRM, orders in the back office, whatever the real source of truth is. If ads report forty and the CRM holds thirty-one, you have found your problem. The fix is deduplication on a transaction or lead ID rather than on a page view.
2. Every event has been promoted to a conversion
In GA4 it is easy to mark events as conversions, and over time somebody usually marks several. Scroll depth, video play, a click on the phone number, the newsletter signup, and the actual purchase all end up in the same column.
The bidding algorithms then optimise toward whichever is most abundant, which is never the one that makes money. You end up buying scroll depth at a very good price.
If everything is a conversion, the platform will find you the cheapest one.
How to check: open your conversion list and read it out loud. Anything that is not money or a genuine step toward money should be an event you can report on, not a conversion you bid toward.
3. Browser and server events are not deduplicated
Running the Meta Pixel alongside the Conversion API is the right call, and it is the standard answer to signal loss. But the two must share an event ID so the platform can recognise that one purchase reported twice is still one purchase.
When the ID is missing or inconsistent, every conversion is counted twice. Reported cost per acquisition halves, the account looks like it has doubled in efficiency, and the optimisation is now being fed noise.
How to check: in Events Manager, look at the deduplication status for your key events. It will tell you directly whether the browser and server events are being matched.
4. Attribution windows are compared as if they were the same
Google, Meta and your analytics platform each attribute differently, over different windows, with different rules for what counts as a touch. Adding their conversion counts together produces a number larger than the number of sales you made.
This is not a bug in any of them. Each is answering a slightly different question. The mistake is treating the answers as interchangeable and then summing them.
How to check: add up platform-reported conversions for last month and compare against actual orders or leads. A gap is expected and healthy. A gap of two or three times means the reporting is being read as if it were accounting, and someone is about to make a budget decision on it.
5. Consent mode is configured to lose data quietly
Consent handling is not optional, and a badly implemented banner does not just reduce your data, it distorts it. If tags are blocked outright rather than run in a restricted mode, you lose modelled conversions entirely, and you lose them unevenly across audiences.
The result is an account that appears to perform worse on some segments purely because those visitors decline more often. Budget then moves away from them, for a reason that has nothing to do with how well they convert.
How to check: look at the share of sessions with consent granted, and whether conversion modelling is active. If your platform reports zero modelled conversions, consent mode is almost certainly not wired correctly.
The order to fix them in
Fix deduplication first, on both the thank-you page and the Meta events. Double-counting is the failure that most directly inflates performance, and it is the one most likely to have justified a budget increase that should not have happened.
Then clean the conversion list, because it determines what the algorithms optimise toward and therefore what you buy for the rest of the year. Attribution alignment and consent mode come after: they change how you read the numbers rather than what the platform does with them.
None of this is glamorous and none of it will show up in a campaign report as a win. It is simply the difference between managing a channel and guessing at one.