The account belongs to a UK-headquartered outdoor and adventure brand with substantial European operations. Annual paid-media spend across Google alone runs at approximately £11.4m; the paid-search and shopping share of that spend is about £7.6m, with the rest going to YouTube and display. The paid lead responsible — call him Sami — has been in the role for four years. He is, by anyone's standard, a competent operator, and this piece is not about his failure to spot the issue. It is about the platform mechanic that produced it and the substantial internal work required to identify and reverse it.
The setup
The account, entering Q4 2025, was running a mature Performance Max stack — six campaigns, segmented by product category, each with its own product feed and its own creative asset group. Alongside PMax, the account maintained a discrete brand search campaign (exact-match, high impression-share target, its own budget line) and a discrete non-brand shopping campaign for the top-margin catalogue segment. This was, by contemporary standards, a well-organised account.
The brand-search campaign was set up in the way most experienced paid leads set up brand search: exact match on the brand name and its common variants, tight negatives, a target impression share in the 90-100% range. It was producing, in Q3 2025, approximately £310,000 of monthly reported revenue at a reported ROAS of 24 — the sort of number brand search always produces, and the sort of number the paid function tends not to interrogate closely because the ratio looks so favourable.
The PMax campaigns had been configured with account-level brand negatives — the standard defensive move, in which the brand's own terms are added as negative keywords at the account level to prevent PMax from bidding on them. This configuration is what most paid leads believe is sufficient to prevent PMax from encroaching on brand search. As Sami discovered, in the specific circumstances of Q4 2025, it wasn't.
What changed in October
In early October 2025, Google rolled out an update to Performance Max's search-partner surface that, on Sami's inspection when he unpicked the issue in December, appeared to have subtly changed the way account-level brand negatives were being enforced across the search partner network. The change was not, so far as anyone can determine, publicly announced. It surfaced in the account's data as a gradual increase in the share of PMax impressions being served against queries that overlapped, or partially overlapped, with the brand's trademarked terms.
The mechanism was as follows. The account-level brand negatives were being enforced correctly on Google's core search inventory. On the search-partner network — where PMax also serves — the enforcement was, apparently, less strict, with a portion of query-match logic delegated to the partner's own retrieval systems. The result was that partial and near-brand queries were being served PMax ads on the search partner surface without the brand campaign being consulted.
From the dashboard's perspective, this looked benign. PMax reported a modest increase in impressions and conversions. Brand search reported a modest decrease in the same. The two numbers moved roughly in opposite directions, in roughly equal magnitude, and the net effect on the account looked neutral.
The net effect on the account was not neutral. It was substantially negative, and the reason took Sami about six weeks to work through.
The real cost
The brand campaign had been producing, in reality, roughly 20% incremental ROAS — a fact Sami had established through a properly-run holdout test in Q1 2025. The remaining 80% of the brand campaign's reported revenue was cannibalised from what would have been organic or direct traffic in the absence of the campaign. In other words, the brand campaign was producing about £62,000 a month of genuine incremental revenue against about £13,000 a month of spend, for an incremental ROAS of about 4.8 — good, but not the 24 the dashboard reported.
The PMax campaigns, on the same holdout methodology, were producing roughly 45% incremental ROAS. Their reported blended ROAS of about 5.2 translated into an incremental ROAS of about 2.3 — meaningfully positive, worth running, but nowhere near what the dashboard suggested.
When PMax began absorbing brand queries on the search partner surface, the reported figures on both campaigns shifted. What Sami's eventual decomposition showed was that PMax was now producing an additional £145,000 a month of reported revenue at an additional cost of £28,000 a month. The reported ROAS of the addition was 5.2. The actual incremental ROAS of the addition was close to zero, because the additional revenue was almost entirely cannibalised from the brand campaign, which was itself only 20% incremental.
Meanwhile, brand search reported approximately £95,000 a month less revenue and £3,500 a month less spend. On the dashboard, this looked like a modest efficiency loss on brand search offset by a modest gain on PMax. In reality, it was a substantial reallocation of low-incrementality brand traffic from a cheap campaign (the brand campaign, at an average CPC of about £0.40) to an expensive one (PMax's brand-blend traffic, at an average effective CPC of about £2.10).
Over the quarter, the account had spent an additional £84,000 on PMax to capture approximately £24,000 a month of genuinely incremental revenue that the brand campaign had previously been capturing for about £3,500 a month. Netted across the quarter, the loss came to approximately £180,000 of avoidable spend. When we then factor in the further Q1 2026 spillover before Sami's remediation took full effect, the total avoidable spend rose to approximately £600,000.
"The dashboard showed two campaigns broadly holding their ground. The account was, in reality, quietly bleeding out. If I hadn't run the incrementality decomposition, I would have found out about it at the annual planning review, when the CFO asked why the paid line looked more expensive than the previous year."
The fix
The fix, once Sami had characterised the issue, was a multi-step remediation that took approximately eight weeks to work through fully.
The first step was to add campaign-level brand exclusions to each of the six PMax campaigns. This was a manual process — the campaign-level brand exclusion feature was, at the time of writing, still not exposed in the standard PMax UI for all accounts, and Sami had to escalate through his Google account team to have it enabled. The escalation took two weeks. Once enabled, campaign-level brand exclusion added a second, more robust layer of defence against the brand-blend issue.
The second step was to restructure the PMax product feeds to exclude the highest-brand-affinity products. The brand's flagship product line was the one PMax was most aggressively targeting, and by removing the flagship SKUs from the PMax feed and moving them to a discrete standard shopping campaign, Sami reduced PMax's incentive to bid on brand queries by a significant margin.
The third step was to expand the brand-search campaign to cover a wider range of near-brand and partial-brand queries. This was, in effect, an admission that the previous configuration had been too tight — the brand campaign had not been defending the near-brand surface, which had left space for PMax to fill.
The fourth step, and the one that has stuck as an ongoing practice, was to institute a monthly incrementality review at the account level. Sami now runs a small, low-cost geo-holdout every month, rotating which campaign is held out, and uses the results to calibrate his incrementality estimates for each line of the account. The monthly practice has, in the six months since introduction, identified two further smaller instances of platform-side leakage that he was able to correct before they cost the account meaningful money.
The lesson
The lesson Sami draws, and the one the Bidfloor editorial team endorses, is that the modern PPC dashboard is not sufficient to detect account-level leakage of the sort that happened here. The individual campaigns' reported numbers were within tolerable bounds. The aggregate account performance was slightly off but well within the noise band of normal quarterly variability. The signal that the leakage was occurring lived in the incrementality decomposition, and the incrementality decomposition is not something the platform will do for you.
Every mature paid function should, in our considered view, have a standing incrementality-monitoring practice — geo-holdouts, MMM triangulation, or (ideally) both — running on a monthly cadence. The cost of the practice is meaningful but bounded. The cost of not running it, in the specific case of this account, was £600,000 over two quarters. That ratio, on our reading, is the case for the practice.