A campaign at 3.4x its cost target turned out to be a tracking bug

This client runs three distinct campaigns with three distinct goals: a lead-magnet download, a community signup, and a booked discovery call. Because all three sat in one account, an earlier framing had treated them as one funnel with one target. They are not the same thing and should never have shared a number.

Judged correctly, two of the three were comfortably beating their targets. The third, the one that actually books revenue-generating calls, was reporting a cost per call more than three times its ceiling.

The reflexive response to a campaign running at 3.4x target is to pause it, cut its budget, or rebuild it. We did none of those, because the first question is always whether the number is real.

It was not. The custom event that should fire when someone books a call was not firing on the site's tracking pixel. Bookings were being recorded under a different event type entirely. That has two consequences, and the second is worse than the first: the reported cost per call was wrong, and the platform's automated bidding had no reliable signal to optimise against, so it was being asked to find booked calls while being shown the wrong events.

The recommendation was therefore to fix the tracking before doing any optimisation work at all. Any budget or bid change made before that fix would have been tuning a campaign against a broken signal, and whatever happened next, good or bad, would have been uninterpretable.

We also separated the three campaigns permanently into their own goals and their own targets, so a passing lead-magnet campaign can never again be averaged together with a failing call campaign into one reassuring blended figure.

The two correctly-tracked campaigns were confirmed as beating their targets, at £2.25 per lead-magnet download against a £3.00 ceiling, and £1.53 per community signup against a £2.00 ceiling.

The third was correctly reclassified from a performance problem to a measurement problem, with a specific, checkable root cause and a fix that had to land before any spend decision was worth making.

This is the least visually impressive case study of the three and it is the one we would point at first. The value delivered was not a better cost per acquisition that month. It was preventing a working account from being torn up on the strength of a number that was never true.