B2B SaaS · AI voice software

Telling a Tracking Break from a Real Decline

Challenge: Analytics fell off a cliff overnight. It was two unrelated problems overlapping: a measurement break, and a genuine organic decline that had started weeks earlier.Results: Arrivals down 19% against captured sessions down 59%. Cause traced to a consent banner blocking tags site-wide. The real decline quantified separately at 42%.Timeline: Diagnosis across two matched 14-day windows

The challenge

Analytics fell off a cliff overnight. The obvious reading was that the site had lost most of its traffic. The obvious reading was half wrong, and acting on it would have sent the team chasing a rankings problem that was not the cause of what they were looking at.

  • Two unrelated events, overlapping. A measurement break on 29 to 30 June, and a genuine organic decline that had started weeks earlier in late May.
  • Conflating them would have misdirected the whole response. Fixing tracking does not recover rankings; chasing rankings does not restore reporting.
  • Three earlier drafts of the diagnosis reached the right conclusion by the wrong mechanism, so the mechanism had to be proven, not assumed.

Approach

Use Search Console as the control. GSC counts clicks at Google's end and is unaffected by anything the site does to its own tags. Comparing it against GA4 over the identical window separates "fewer people arrived" from "we stopped counting the people who arrived".

Compute a capture rate. GA4 sessions per Google click, split by new and returning. This is the single number that exposes a tag problem: if arrivals are steady and captured sessions collapse, the loss is in the measurement layer.

Find the mechanism in the page source, not in the dashboard. A diff of the deployed HTML showed the consent-management script had been swapped from a disabled 45-byte stub to a fully armed 452,838-byte build in opt-in mode with auto-blocking on, moved into the head. It was blocking every tag site-wide before consent, which most visitors never gave.

Then quantify the real decline separately, once the measurement noise was removed from the picture, and attribute it to specific pages.

Make every figure reproducible. A verify script rebuilds each number from the raw exports, so the client can re-run the analysis rather than take it on trust.

Results

  • -19% against -59% — GSC clicks versus GA4 organic sessions over the same window. Arrivals barely moved; captured sessions collapsed.
  • Capture rate for new users fell from 0.89 to 0.14 GA4 sessions per Google click. Returning users held at 0.73, because their consent cookie already existed.
  • The real, separate decline was -42% — GSC clicks 11,368 to 6,647, impressions down 53%, over two matched 14-day windows.
  • 82% of the click loss came from the homepage alone, which moved from position 12.2 to 16.7, focusing the recovery work on one template.

Key insight

A consent banner is analytics infrastructure. This one was swapped from disabled to fully armed in a routine deploy and silently stopped almost all tag firing, while Search Console showed arrivals were roughly flat. On one matched weekday GSC recorded 610 clicks against 615 the week before, while GA4 new organic sessions went from 560 to 89. Without the control dataset, the team would have been recovering rankings that had not moved and ignoring the homepage decline that had.