A conversion tracking audit is the check that your reported numbers describe reality, and the order you run it in decides how long it takes. Start with the tags and you will spend a day. Start with the stack and most faults surface inside an hour.
Fourteen checks below, in the sequence I actually use. None of them need developer access to perform, though several need one to fix.
This is for you if you inherited an account, or if a number stopped making sense and you want a method rather than a hunch.
Start the conversion tracking audit with the stack
Establish what the site is built on and what is doing the recording before you look at a single tag.
WordPress with a plugin. Shopify with an app. Hand-written code in a template. A tag manager sitting over the top, possibly alongside one of the others.
That answer changes where every subsequent fault will be. A duplicate conversion in a plugin-driven setup lives somewhere completely different from one in a tag manager, and searching the wrong place is how an audit turns into an afternoon.
It is also quicker to establish than people expect. View the page source, look at what is loading, and ask whoever built the site one question.
Then work backwards from the number you distrust to the thing that fires it. Backwards from the reported figure, not forwards from a setup guide. That single discipline is what makes this fast.
Checks one to four: what is actually counting
The first four questions establish whether your conversion actions are what you think they are.
One. List every conversion action and write down what triggers each one. If two describe the same moment, you have found duplication before doing anything else.
Two. Check the count setting on each lead and call action. Count every occurrence is almost always wrong for lead generation, and it is set wrongly more often than not.
Three. Check whether a submit click, a call button click and a thank-you page load are all counting separately. One person can generate three conversions that way.
Four. Try loading your thank-you page directly, by typing the URL. If it fires a conversion, anybody who bookmarks it converts on every visit, and so does every internal person testing the site.
Those four take about fifteen minutes and they find the largest single category of fault.
Checks five to eight: are the values real
The plausibility checks, and this is where the most damaging faults live.
Five. Divide reported conversion value by reported conversions, per campaign rather than for the account. Ask whether that average order or enquiry could exist in your business.
Six. Walk the funnel and confirm each step is smaller than the one before. More purchases than checkouts is impossible.
Seven. Check for campaigns with values wildly out of line with the rest. One campaign at $8 an order where everything else sits in the thousands is a broken value, not a bargain.
Eight. Compare your platform conversion count against your CRM for one matched month. Write the gap down as a percentage, because that baseline is what tells you next time whether something moved.
Check five is the one I would keep if I could only keep one. It found a fault on an account reporting a return of 44.37, where a single campaign recorded 11 purchases against $86.35 of value.
Worth knowing why this check earns its place above more sophisticated ones. Gordon, Zettelmeyer, Bhargava and Chapsky compared observational attribution against randomised experiments across 15 Facebook experiments and 500 million user-experiment observations, finding estimates off by a factor of three in half the studies. Sophistication does not protect you from a wrong answer. A plausibility check does, because an implausible number is wrong regardless of how it was produced.
Checks nine to eleven: what is reaching the platforms
Now the connections, which fail silently and after unrelated work.
Nine. Submit a real test enquiry through a live advert click, and confirm it arrives in your CRM with the click identifier attached. This is the single most informative five minutes in the whole conversion tracking audit.
Ten. On Meta, open Events Manager and check your money event shows a deduplication status rather than two separate counts. Two counts means it is double counting right now.
Eleven. Check which conversion actions are imported into Google Ads and which are set as primary, because that set is what your bidding is chasing. Accounts frequently have events marked important in analytics that do nothing in advertising.
Checks twelve to fourteen: what the reporting says
The last three are about interpretation rather than plumbing.
Twelve. Add the attribution setting column on Meta and confirm your campaigns are not running a mixture you never chose. Comparing campaigns measured on different bases is comparing nothing.
Thirteen. Check what share of reported conversions are view through, per campaign type. On display and video that can be most of the total, and it belongs in its own line rather than blended into the headline.
Fourteen. Establish whether consent modelling is running at all, which needs 700 ad clicks in seven days per country. Below that threshold, refused consent is simply lost.
What to do with what you find
Fix in order of how much damage each fault causes, which is not the order they were found in.
Counting faults first. Duplication, wrong values and spam entering the signal all corrupt both your reporting and your optimisation, so they cost twice.
Attribution and reporting issues second. Those affect how you read the numbers rather than what the algorithm learns.
And warn people before you fix anything. Correcting duplication makes conversions drop and cost per lead rise, visibly, in a week when nothing about performance changed. Explaining that in advance is a competence conversation. Explaining it afterwards is a different one.
When the conversion tracking audit is finished
There is a clean test for completion and it is not a checklist.
You should be able to say, in one sentence each, what every conversion action counts and where every reported number comes from. Not roughly. Specifically.
If any part of that sentence is a guess, the guess is your next check. That standard is higher than most accounts ever reach, and reaching it once makes every future audit take twenty minutes rather than an hour.
Write the answers down while you have them. The document you produce is worth more than the faults you fixed, because it is what lets somebody else check this next time.
What to change this week
Three steps.
Run checks one to eight, which need nothing but the reporting interfaces and about half an hour.
Then do check nine, the live test enquiry, because it is the highest information per minute in the entire process.
Then write down what every conversion action counts, in one line each, and keep that document somewhere the next person will find it.
What conversion numbers actually measure is the reasoning behind all of these checks, conversion event deduplication covers the most common fault, attribution windows covers check twelve in detail, and the twelve checks I run on any account is the wider version that includes the campaign work. The case studies show what the numbers look like afterwards. If you would rather somebody else ran this on your account, book a teardown.