When "CPL fell 87%" falls apart in 30 minutes.
We spent last week rewriting every case-page on this site. We had 19 of them, some written years ago by a previous agency owner, all of them claiming impressive deltas: +1,700% conversions, +212% conversion rate, −87.5% cost per conversion. The kind of numbers that look great in a pitch deck and uncomfortable when a sceptical prospect asks where they came from.
So we did the uncomfortable thing. We pulled the original Google Ads screenshots, opened the engagement timelines, and reconciled the live page numbers against the source data. The result: most of them didn’t survive 30 minutes of scrutiny.
The shapes of inflation
There were three recurring patterns. Once you’ve seen them, you can’t unsee them.
1. Math against zero
A real example from one of our own pages: “0 → 8,000 conversions” means an infinite improvement. Technically true. Practically meaningless. The before-state was “we hadn’t set up conversion tracking yet” — not “the campaigns were producing zero leads”.
The honest version of that case is much less spectacular and much more useful: we set up proxy conversion tracking for a distribution business that doesn’t transact on-site, attached a €0.40 value to the soft engagement events that correlate with downstream dealer-sales, and watched Google’s bidder use those signals to find better placements over 14 months.
2. Windows that don’t reconcile
An ex-agency case claimed “+1,700% conversions, 5 to 90 per week”. The supporting screenshot was a 26-week window with 652 total conversions. 652 divided by 26 is 25 conversions per week, not 90. Both numbers were probably true at some point — the 90/week was likely a peak month, the 25/week was the blended average across the window the screenshot covered — but the page presented them as the same thing.
3. KPIs from systems that weren’t measuring
The most common pattern, and the hardest to catch from a single screenshot: revenue claims on a screenshot that doesn’t have a revenue column. “+303% revenue” on a Google Ads overview that only shows clicks and conversions. The number might be real — the client may have told the agency that revenue tripled — but it’s not backed by the screenshot anyone is being asked to trust.
What we kept
After the cull, the 19 pages now lead with numbers that meet three tests:
- The screenshot or live API data exists, and we can show it (anonymised where the engagement requires).
- The window in the screenshot matches the window in the headline, or the page explicitly tells you the screenshot is wider than the engagement and uses trajectory language instead of blended numbers.
- The before-state is actually the before-state, not the “before we measured anything” state.
It’s a less spectacular portfolio. The hero numbers shrunk. −72% blended CPL with a trajectory-climb is harder to fit in a tweet than −87.5% cost per conversion. But the smaller numbers survive the 30-minute reconciliation that a serious prospect will do before signing a retainer, and the bigger numbers don’t. That’s the trade we made.
The audit version of this exercise
The same exercise runs against most account audits we do. The first thing we look for isn’t whether the campaigns are good. It’s whether the conversions being measured are actually leads, the windows the dashboards report match what the screenshots show, and the “before” numbers the agency is comparing against were ever real.
That part of the audit is free and usually unpleasant. We do it before any engagement.