Five Tracking Failures Nobody Noticed

None of these broke anything you could see. The pages loaded, the forms said thank you, the reports had numbers in them. Each one quietly made those numbers wrong, or lost a person who wanted to get in touch.

The setup. Four consumer health brands of a European pharmaceutical group, July to October 2026. I worked through a digital agency, next to the group's CTO, the brands' marketing team and their ad agency. Part of the work started from the ad agency's tracking audit: I went through it slide by slide and answered every point, including the one where they were right and I was not.

1. A cookie banner that had to really mean no

Data sent to Meta from the server had to depend on the visitor's answer to the cookie banner. If someone clicks reject, nothing should go to analytics or ads. It is easy to assume that is what happens.

I checked it the hard way: a wiped browser profile, consent denied, every request the page made counted one by one.

On the way I caught a problem in the consent check itself. It asked the banner for its language file once a second, forever, on every open page. Nobody would ever notice that by looking at the site. I changed it to decide once per page, so the page no longer pays for a request that could never succeed.

2. Two pixels, every visit counted twice

This brand has no shop of its own. Its conversion is the click to a retailer, so that click is the number the ads are judged by.

Every page view went to two Meta pixels. That means doubled counts in one account, or the brand's data going to an account it does not control. I confirmed with the client which pixel was theirs, paused the other one, and then measured on the live site that a retailer click now reports to one account only.

3. A €1,140 order that never happened

The tag manager held old tags that were still set up with the documentation's example purchase: order id "your order id", 1,140 euro. They were quiet only because their triggers happened to be broken.

That is a trap. The day someone fixes a trigger for an unrelated reason, a fake €1,140 sale lands in the reports and in the ad account. I paused those tags, so nothing depends on a broken trigger staying broken. Paused, not deleted, so every change can be undone.

4. Analytics that recorded nothing

On one live site, the product analytics recorded nothing at all. The cause was in the build: the site was built without its analytics keys, so the tracking code loaded and quietly did nothing.

Untangling it, I mapped three separate analytics instances and which site reported where, then fixed the configuration.

5. Forms that said thank you and lost the request

Doctors could request product samples through a form. The old forms ran through a marketing tool and attracted spam, so they were rebuilt as forms the website renders itself, sending straight to the new CRM. The new forms looked fine. Then I tested them end to end, with the mail redirected to my own inbox.

The anti-spam timer threw away any form sent before the page was fully ready, while the visitor still saw a thank-you. A fast, decisive visitor was exactly the one who got lost. I fixed it, tested both forms again until each delivered its email and created a complete lead in the CRM, and then removed every test contact so the client's CRM held no traces of my testing. I also checked that the rate limit sees real visitor addresses, not the address of the proxy in front of the site.

Because the form was about to move between systems, I built the sample request conversion into the analytics itself, so it keeps counting when the form changes. I verified it without creating a single live lead.

And one near miss

On another brand, the client was about to switch off the old marketing tool and asked what still depended on it. My inventory showed that almost nothing in it had ever run: two emails sent in its whole history. My first conclusion was that its contact form was dead too.

I re-checked before reporting, and found that form embedded on all four product pages. 71 people had used it. It became the thing that had to be replaced before the switch-off, instead of being deleted with the rest.

Check your own site in 15 minutes

You do not need a developer for the first pass:

1. Count your pixels. Install Meta Pixel Helper, open your site, and look at how many pixel IDs it lists. You want one, and you want to know it is yours.
2. Send your own form, fast. Fill it in right after the page opens, the way someone in a hurry would. Check that the email arrives and that the lead appears wherever leads should go.
3. Ask for the list of conversion tags. Anything named test or example, or carrying a sample value, should be paused.
4. Reject cookies and ask what still goes out. Whoever runs your tracking should be able to show you, not just tell you.