Four ways the number gets wrong, and none of them look broken

Why conversion tracking is wrong card showing three times, the factor by which attribution estimates missed a randomised benchmark
Contents 8 sections

Most of the ad accounts I open are reporting numbers that are wrong somewhere, and almost none of them look it. That is the awkward part of understanding why conversion tracking is wrong so often: broken tracking does not announce itself, it just reports something plausible and lets you act on it.

The short answer is that the failure is silent by design. A tag that stops firing, an event counted twice, a value passed as zero, all produce a dashboard that looks exactly like a working one.

This is for you if your reports and your bank balance tell different stories.

Why conversion tracking is wrong more often than it is right

There is no error state. That is the whole problem in one sentence.

When a payment fails, somebody finds out immediately. When a conversion tag fails, the number simply gets smaller, and a smaller number in a noisy weekly report is indistinguishable from a slow week.

So faults survive. They survive for months, sometimes for the life of the account, because nothing in the system is designed to tell you that what it is reporting is not what happened.

Add to that how many hands touch the chain. A developer changes the checkout. A marketer adds a tag. An agency sets up a second conversion action. None of them see the whole path, and the fault appears in the gap between them.

The four ways conversion tracking goes wrong

Every tracking fault I have found sits in one of four buckets.

Under-counting. A conversion happens and never gets recorded. A tag did not fire, an ad blocker stopped it, or consent was refused. This is the honest failure, because it makes performance look worse than it is, so somebody investigates.

Double counting. The same action gets recorded more than once. Two tags on one page, or a conversion set to count every occurrence instead of one.

Misattribution. The conversion is real and gets credited to the wrong thing. Paid gets credit for organic, or one campaign absorbs another’s results.

Wrong values. The count is right and the money attached is nonsense. This is the least discussed and the most damaging, because every return figure you calculate rests on it.

Notice what three of those four have in common. They make the account look better, not worse.

Why nobody notices, and why that is predictable

A fault that flatters an account is a fault with a protector.

The campaign that appears to be performing best gets more budget. It also gets less scrutiny, because nobody audits the thing that is working. So the error grows in proportion to the money it attracted, which is exactly backwards from how you would want it to behave.

I have watched an account move budget towards its “best” campaign for months, where best meant most broken. Nobody did anything unreasonable at any point. Each decision was correct given the number in front of them, and the number was wrong.

That is the real reason why conversion tracking is wrong in so many accounts and stays wrong. The incentive to look is weakest exactly where the problem is worst.

What the research measured

This is not just a practitioner’s grumble, and the best evidence is unusually direct.

Gordon, Zettelmeyer, Bhargava and Chapsky compared observational attribution against randomised experiments across 15 Facebook experiments covering 500 million user-experiment observations. In half the studies, the estimated effect was off by a factor of three.

One checkout study is worth stating on its own. The observational method reported a 1,306% lift. The randomised benchmark was 2.4%.

That is not a rounding problem or a settings problem. It is a method producing an answer of the wrong order of magnitude, at a sample size no agency will ever match. The data is from 2018 and nothing about the mechanism has changed since.

Keep that beside every confident percentage you are shown, including your own.

The ten minute check for conversion tracking that is wrong

Two tests, and between them they catch the majority of real faults.

First, the plausibility check. Take reported conversion value and divide it by reported conversions. Ask whether that average order or enquiry could exist in your business. Implausible averages are the fastest possible sign that values are broken, and they show up per campaign rather than only in the account total.

Second, the direction check. Walk the funnel and confirm each step is smaller than the one before it. More purchases than checkouts is impossible. Attribution windows can make funnels read slightly out of order, so treat a small inversion as a question rather than a verdict, and a large one as a fault.

Neither needs a developer. Both can be done from the reporting interface in a coffee break.

What this changes about how you read a report

Stop treating the number as a fact and start treating it as a claim.

A claim can be checked, and it is usually checked in about a minute by asking whether it could be true. Most people never ask, because the figure arrives in a dashboard rather than in a sentence, and dashboards feel like measurements rather than opinions.

The habit worth building is small. Every time a figure surprises you in a good way, check it before you celebrate. Nearly all of the tracking faults I have found started with somebody being pleased.

Who owns this, and why that is the real reason conversion tracking is wrong

Almost every fault I find lives in a gap between two people who each did their job properly.

A developer changed the checkout, with no reason to know a marketing tag depended on the old button. The marketer added a conversion action and could not see that another one already counted the same thing. Then an agency inherited the account, three years of history deep, with nothing written down.

Nobody was careless. The chain simply crosses more than one person’s responsibility, and nothing in any of their tools shows them the whole path.

That is why the technical fixes in this article matter less than one organisational one: somebody has to own the number. Not own the campaigns, not own the website, own the number. One named person whose job includes noticing when it stops making sense.

On accounts where that person exists, faults get caught in weeks. Where nobody owns it, I have found problems that were years old, sitting in plain sight in a dashboard several people looked at every Monday.

It is not a technology problem at that point. It is a question of who is expected to look.

What to change this week

Three steps.

Run the plausibility check on every campaign, not just the account total. One broken campaign in a good account is the common shape, and account-level numbers hide it.

Then write down which single system the business treats as the source of truth for revenue. Most arguments between marketing and finance are two people quoting two tools.

Then put a check in your calendar for the week after any site change, because rebuilds cause most new faults and nobody remembers to look afterwards.

What conversion numbers actually measure covers the mechanism behind all four failure modes, and why your reported cost per lead is wrong goes deeper on the misattribution half. Organic gets miscounted in the same ways, which is worth knowing before you compare channels, and my approach to SEO treats measurement as part of the work rather than an afterthought. If you would like somebody to run these checks against your own account, book a teardown.

Frequently asked questions

Why is conversion tracking wrong so often?

Because it fails silently. A missing tag, a duplicated event or a wrong value all produce numbers that look completely normal in a dashboard. Nothing turns red. The only way to find these faults is to go looking for them, and almost nobody does until something forces the issue.

What are the main ways conversion tracking goes wrong?

Four: conversions that never get counted, conversions counted more than once, conversions credited to the wrong source, and conversions carrying the wrong value. The first looks like poor performance. The other three look like good performance, which is why they survive so much longer.

How wrong can the numbers actually be?

Very. A published comparison of observational attribution against randomised experiments found estimates off by a factor of three in half the studies tested. In one case an observational method reported a 1,306% lift where the controlled benchmark was 2.4%.

Does this mean I should ignore platform numbers?

No, use them for steering rather than for truth. A figure that is wrong in a consistent direction still tells you which campaign is improving week to week. What it cannot do is tell you how much profit your advertising actually caused.

What is the quickest way to spot a fault?

Divide reported conversion value by reported conversions and ask whether that average is plausible for your business. Then check your funnel counts in the right direction. Those two checks take about ten minutes and catch the majority of real faults.

Why does nobody notice these problems?

Because three of the four failure modes flatter the account. A campaign that appears to be performing well gets more budget and less scrutiny, so the fault grows along with the spend it earned. Only under-counting causes anyone to investigate.

Is conversion tracking less often wrong on bigger accounts?

Not in my experience. Bigger accounts have more tags, more integrations, more people touching the site and more history, so there are more places for something to break quietly. Complexity adds failure modes faster than budget adds scrutiny.

How do I stop this happening again?

Check after every site change without exception, because rebuilds cause most new faults. Then run the plausibility check monthly. Neither takes long, and the alternative is discovering the problem when somebody compares your reports against the bank account.

Want a second pair of eyes on your account?

Book a free teardown

Book a free teardown

Send me your worst-performing campaign.

I will tell you what is wrong on the first call. No obligation, no hard sell, and I will say honestly if I am not the right fit.

Book a free 20-minute account teardown

Replies within one business day, Lahore time.