Stop measuring conversions on the thank-you page alone

The standard conversion setup fires on the page someone reaches after succeeding. It is easy, it is reliable, and it measures exactly one thing: how many people got through.

It cannot tell you anything about the people who did not, which is the population you can actually do something about.

The three events worth having

form_start — fired when someone focuses the first field. This is intent, and it is the denominator that makes everything else meaningful.

form_error — fired when validation fails, with the field name as a parameter. This is the single most actionable analytics event most sites do not have.

form_submit — the success, as now.

With those three you can compute the numbers that matter: what proportion of people who started finished, and which field is losing them.

What it reveals

Typical findings, all common, none visible from a thank-you page count:

A required phone number halving completions on mobile.

A postcode validator rejecting valid formats — usually spaces, or Northern Irish and Channel Island codes — which shows up as one field generating most errors.

An error state that clears the form, so a single mistake costs everything typed. The signature is a spike in form_error followed by no form_submit and no second attempt.

A submit button outside the form element, which does nothing at all on some browsers. form_start fires, form_submit never does, and there are no errors, because nothing was ever validated.

That last one sounds implausible until you find it. Then you find it again somewhere else.

Implementation without a developer

Tag Manager can do all three without touching the site: a trigger on form field focus, one on the visibility of an error element, and the existing submit trigger. It is imperfect — CSS-selector-based triggers break when the markup changes — but it is available today and it is dramatically better than nothing.

The durable version is a developer firing explicit events from the form's own code, where the error state is actually known rather than inferred.

The reporting shift

Once you have starts and errors, the conversion question changes from "how many enquiries did we get" to "of the people who tried to enquire, how many succeeded, and what stopped the rest".

The second question has answers you can act on this week.