Trial CPA from €81 to €47: the PMax overlap teardown

What was broken

On paper, PMax delivered trials at €52 while search delivered them at €89, so budget kept migrating to PMax. The audit asked one question the reporting never had: where would those PMax conversions have gone without it? No brand exclusions existed. Brand searches flowed into PMax, converted as they always would, and were counted as PMax wins. Meanwhile exact-match impression share on the core non-brand terms had slid noticeably in six months as PMax outbid its sibling campaigns internally.

The blended effect flattered the account while starving the part of it that wins strangers. Trial CPA on genuinely new demand, reconstructed from search-path data, sat near €81. That was the real number the account was paying, against a business-services median CPL of $93.69 for far colder conversions than a free trial.

The fixes, in order

1. Weeks 1–2: brand exclusions applied to PMax; a dedicated exact-match brand campaign took ownership of name queries with a capped budget.
2. Weeks 2–5: non-brand search rebuilt into intent tiers around the queries PMax had been skimming; negatives per the themed-list system, 198 added in the first month.
3. Weeks 5–8: PMax relaunched at 15% of budget under the capped test protocol, URL exclusions on careers and support paths, judged only on incremental trials.
4. Weeks 8–12: budget rebalanced from verdicts, with PMax settling at 62% less budget than its pre-fix peak and search carrying the recovered core terms.

What the channel report showed

The diagnosis would have taken a fortnight of inference two years ago. Since November 2025, every account carries PMax's channel performance report, which itemizes where the money went, surface by surface, and reads out search terms with the same detail Search campaigns have always offered. Opening it on this account made the case in one screen: the "star" campaign's conversions sat overwhelmingly in Search placements, and its search terms read like the client's brand book. Prospecting spend, meanwhile, pooled in Display placements whose clicks never reached a trial.

The same report became the test's referee at relaunch. The capped PMax's monthly channel read is a standing item in the change log, and the +11-point recovery in exact-match impression share on core terms is corroborated there by the near-disappearance of brand queries from PMax's own search-terms view. When a tool finally lets you check the homework, checking it monthly is the least an account deserves.

What we left alone

The trial funnel and the pricing page, both regular requests for "quick CRO wins." Trial signups converted to paid at a stable rate throughout, meaning the funnel was doing its job once honest traffic reached it; the €81 problem lived entirely in what the ad account paid for attention. We also kept YouTube out of the relaunch scope despite decent demo-video assets: with the CPA target freshly met, the account earned a quarter of stability before the next experiment, and the log says exactly that.

The lift, with context

Trial CPA on new-demand traffic finished at €47, down 42%, while weekly trial volume held within 8% of the pre-fix average. The recovered exact-match impression share, up eleven points on core terms, explains most of it: the account stopped paying PMax premiums for demand its own search campaigns could buy cheaper. Ines Beltrán flags this engagement internally as the cleanest argument for never reading PMax reporting without asking the counterfactual question first.

What this teardown generalizes to

Any account where PMax launched without brand exclusions is carrying some version of this overlap, and the flattered reporting hides it by design. The before-and-after check takes an afternoon inside the 90-point audit, where PMax cannibalization is one of the six standing sections. AI Max deserves the same skepticism as it rolls out: test capped, judge on increments, write the verdict down.