Two of the ecommerce accounts I run report a return on ad spend of 1.58x and 44.37x. Same platform, same marketer, overlapping months in 2026. One of those numbers is real, the other is a measurement fault, and conversion tracking is the only thing standing between them.
That is the short version of why any of this matters. Every decision you make about paid media rests on a number that some code somewhere decided to record. If the code is wrong, the decision is wrong, and nothing in the interface will tell you.
This is for you if you spend money on ads and have never checked what your conversions are actually counting. If you cannot track conversions accurately, everything downstream of them is guesswork wearing a decimal point.
What conversion tracking actually measures
It records that something you care about happened, then ties it back to the advert that caused it.
That is two jobs, not one. The recording part is usually fine. The tying-back part is where almost every problem lives.
A form submission fires an event. The event carries an identifier saying which click brought this person here. The platform matches the two and credits the advert. Break any link in that chain and you still get numbers, just wrong ones.
The load-bearing word is “caused”. A conversion count is a claim about cause, and cause is genuinely hard to establish. Most of what follows is about the ways that claim goes wrong.
Two accounts, two numbers, and one of them is fiction
Both are ecommerce, both on Meta, both mine, over overlapping periods.
The first is a US flag and banner business. Across its six sales ad sets it returned 1.58x, from 64 purchases at $67.24 each. Strip out one failed geographic test and it is 1.99x from 61 purchases at $53.72. Those are ordinary, believable numbers.
The second is a US automotive equipment business. It reports 44.37x.
Nobody returns 44.37x. If they did they would put every penny they own into it and never speak to a marketer again. That figure is an artefact, and I can show you where it comes from.
The contrast is the useful part. Same person, same platform, same months, and the reported results differ by a factor of twenty-eight. That gap is not skill. It is conversion tracking.
How I know which number is broken
The funnel gives it away before the arithmetic does.
The automotive account records 542 adds to cart, 374 checkouts started, 21 additions of payment details, then 193 purchases. More people bought than entered a card, which cannot happen.
Attribution windows explain some of that. A purchase can be credited to a click from weeks earlier while the add to cart sits in a different window, so funnels do sometimes read out of order. That much is normal, and it is why funnel order on its own is not proof of a conversion tracking fault.
Here is what windows cannot explain. One campaign in that account records 11 purchases against $86.35 of total order value. That is roughly $8 an order. Every other campaign in the same account sits between $3,000 and $7,900.
An $8 order does not exist in a business selling automotive equipment. Something in the values being passed back is broken, and once the values are wrong, every figure built on them is wrong too.
That is one campaign, and it is enough. You do not need to audit everything to know a number is untrustworthy. You need one thing that cannot be true.
Why conversion tracking breaks without anyone noticing
A broken conversion tracking setup does not throw an error. It reports confidently.
Three ways I see it happen, in order of how often:
A site rebuild. Someone redesigns the checkout, the event still fires, and a parameter it used to send quietly stops arriving. Nothing looks broken from the outside.
Duplicate counting. Two tags record the same action, or a conversion is set to count every occurrence instead of one, or a button click and a thank-you page load each count separately. The number goes up and everybody is pleased.
A value that is wrong rather than missing. This is the automotive case and it is the worst kind, because an absent conversion looks like a problem while an incorrect one looks like a result.
What the research says about trusting a reported number
My accounts are not the only evidence, and the academic work here is unusually clear.
Gordon, Zettelmeyer, Bhargava and Chapsky compared observational attribution against randomised experiments across 15 Facebook experiments and 500 million user-experiment observations. In half the studies, the estimated effect was off by a factor of three. In one checkout study the observational method produced a 1,306% lift where the randomised benchmark was 2.4%.
That is a comparison at a scale no agency can match, and it says conversion tracking does not measure what most people assume it measures. The data is from 2018 and the mechanism has not changed since.
None of that makes the platforms dishonest. Ad platform reporting is built to credit the platform, using only what the platform can see. That is what the tool is for, and it is why a second opinion matters.
More recently, Seer Interactive tested Meta’s incremental attribution setting across $1.05 million of ad spend on six accounts in April 2025. Meta reported 87% of conversions as incremental. On the same accounts, GA4 put it at 67%. Twenty percentage points apart, from two systems watching the same events.
What conversion tracking cannot tell you
Worth being honest about the ceiling, because perfect measurement is not the goal and is not available.
Wired correctly, it tells you what was recorded and credited. It does not tell you what would have happened anyway. Somebody who was always going to buy, who saw your advert, clicked it and bought, is recorded as a conversion you caused. Often you did not.
Closing that gap is what incrementality testing is for, and that is a different discipline with a different price tag.
So the goal is not truth. It is a number that is wrong in the same direction every month, because you can steer with that. A number that is wrong in a different way each month is worse than useless, and that is what a broken setup gives you.
Where to start if your conversion tracking is a mess
Stack first, then tags. That order saves hours.
Establish what the site is built on and what is doing the recording. WordPress with a plugin, Shopify with an app, hand-written code, or a tag manager. The answer changes where you look, and it is quicker to find than people expect. Once you know the stack, you can see where the conversion data is produced rather than guessing.
Then work backwards from the number you distrust to the thing that fires it. Not forwards from a setup guide. Backwards from the reported figure.
Most faults surface on that walk. An event fires twice. A value is hardcoded. The click identifier never reaches the form. Somebody bookmarked the thank-you page, so they convert again on every visit.
What to change this week
Three things, and the first takes ten minutes.
Take your best-performing campaign and check whether its numbers could be true. Divide reported value by reported conversions and ask whether that is a plausible order or enquiry for your business. That single check is what found the $8 order.
Then look at whether your funnel counts in the right direction. More purchases than checkouts means something is wrong, whatever your attribution settings say.
Then agree which one number the business will judge paid media on, and which system it comes from. Most arguments between marketing and finance are two people quoting two tools and each assuming the other is wrong.
The twelve checks I run on any account is the practical version of all this, why your reported cost per lead is wrong covers the attribution half in detail, and sending closed-won data back to the platform is the single change that makes these numbers start meaning something. If you would rather this were set up properly than explained, Google Ads management includes it, and you can book a teardown.