The 71% Conversion Drop That Never Happened

A demo conversion number fell from roughly 407 to 117 month over month. On a dashboard, that reads as a 71% collapse — the kind of chart that starts a Slack thread, then a meeting, then a “what changed with the website” investigation.

Nothing had changed with the website. The demos were fine. What had changed was the definition of a conversion — and that is a very different problem to solve.

Here is the full diagnosis, because the trap this fell into is one almost every RevOps, demand gen, and analytics team hits eventually: a config change quietly rewrote what the metric meant, and the number moved for a reason that had nothing to do with the business.

The symptom: a funnel cliff with no cause

The alarm came in through a funnel task. Demo conversions had dropped hard between two 28-day windows, and the drop was steep enough to look structural rather than seasonal. When a number falls that far that fast, the instinct is to go hunting for a broken form, a busted tag, or a landing page regression.

I did none of that first. When a conversion count drops off a cliff, the single most common cause isn’t lost conversions — it’s a change to what counts as a conversion. So before touching a form or a tag, I went to the key event configuration in GA4.

The dead end I skipped on purpose

It would have been easy to burn a day here. The tempting path is to assume the funnel itself broke: check whether demo_submit is still firing, replay the form, inspect the GTM tag, diff the landing page. All of that is real work, and all of it would have found nothing wrong, because nothing was wrong with the funnel.

The faster path is to ask a cheaper question first: did the meaning of the metric change during the window I’m comparing? A definition change is invisible on the conversion chart itself — the line just moves — but it is trivial to confirm in the event settings. Check the cheap explanation before the expensive one.

The root cause: a form view wearing a conversion’s badge

In GA4, key events are what most teams still think of as conversions. The reporting hangs off them, and anything marked as a key event gets counted as a goal completion.

Sometime earlier in the quarter, a teammate had marked demo_form_view as a key event. That event fires when someone sees the demo form — not when they submit it. For the weeks it stayed flagged, every form view was counted as a conversion. The baseline conversion number was inflated by a large volume of people who looked at a form and did nothing.

Mid-quarter, someone correctly removed demo_form_view from key events. The moment it stopped counting, the inflated baseline deflated. On the month-over-month chart, that deflation looked exactly like a cliff — because the “before” window was padded with form views and the “after” window wasn’t.

The 71% drop wasn’t demos disappearing. It was a vanity metric being switched off, and the switch happened to fall inside the comparison window.

The real picture, once the noise was removed

When I isolated the events that actually represent intent — form submissions and trial registrations — the story reversed completely:

• demo_submit: up roughly 20% period over period
• trial_registration: up roughly 45% period over period

Both real conversion events were growing. The business was healthier than the prior period, not weaker. The only thing that had “dropped” was a number that never should have been counted as a conversion in the first place.

Why this matters beyond one dashboard

A form view is an engagement signal. A form submission is intent. Collapsing the two into a single “conversion” number does two kinds of damage at once.

First, it inflates your conversion rate. When form views count as conversions, every optimization looks like it works and every real decline gets masked. You lose the ability to tell whether the funnel is actually improving.

Second — and this is the expensive one — it manufactures false alarms. The day someone corrects the mistake, the metric craters, and a team spends real hours investigating a business problem that doesn’t exist. The correction looks identical to a collapse. That’s the part that burns time and credibility.

For a RevOps or demand gen team, a fake conversion drop is worse than no data, because it moves people. It reallocates budget, it reprioritizes sprints, it gets a healthy channel flagged as broken.

The fix

The fix here was mostly a decision, not a deployment:

1. Leave demo_form_view out of key events. Form views are engagement, not conversions. They belong in an engagement report, not a goal count.
2. Keep key events limited to real intent — demo_submit and trial_registration. Those are the events tied to pipeline, and they’re the only ones that should drive conversion reporting.
3. Close the investigation with the actual root cause on record, so the next person who sees the “drop” in a historical chart knows it was a definition change and not a business event.

No form was broken. No tag needed rebuilding. The right output was an accurate explanation and a cleaner key-event list.

The reusable lesson

If you pull one thing from this, make it the diagnostic order.

When a conversion metric moves sharply, check the definition before you check the funnel. A change to what counts as a conversion is invisible on the conversion chart and is the most common cause of a sudden cliff. It’s also the cheapest thing to rule out. Confirm the metric still means what it meant across your whole comparison window before you go looking for a broken form.

Three habits make this a non-event instead of a fire drill:

• Never mark engagement signals as key events. Views, scrolls, and clicks are not conversions. If it doesn’t represent intent, keep it out of the goal count.
• Cross-check the composite against the components. Before trusting a single conversion number, look at the underlying events independently. If the composite dropped but demo_submit and trial_registration are up, the composite is lying to you.
• Annotate key-event changes. GA4 supports annotations for a reason. A one-line note the day a key event changes turns a future “why did this drop” mystery into a five-second answer.

The most dangerous number on a dashboard isn’t the one that’s wrong. It’s the one that’s wrong in a way that looks like a business emergency. This was one of those — and the fix was to stop counting the thing that never should have been counted.