The brand is a mid-market UK financial-services company, running roughly £1.25m of annual programmatic display spend across DSPs, primarily through open exchange and a handful of PMPs. The head of programmatic — call him Timo — inherited the account when he started in the role in late 2024. His remit, on paper, was to modernise the programmatic function. His actual first-quarter task, once he had started reading the delivery reports, became something much more specific and considerably less comfortable: figure out how much of the account's spend was going to MFA sites, and then figure out what to do about it.
The audit
Timo's approach, worked out over the first six weeks in role, was structured but not particularly technical. He pulled twelve months of delivery data at the domain level, exported from the DSP, joined against three independent MFA classification data sets (one from a large third-party verification vendor, one from the DSP's own quality-scoring system, one from an open-source list maintained by an industry researcher). The three sources agreed, on the majority of domains, in their classifications. Where they disagreed, Timo manually visited the domain and made a judgement call.
The full audit covered 4,142 domains that had received meaningful delivery in the twelve-month window. Of those, 812 were classified as MFA by all three sources; a further 340 were classified as MFA by two of three; a further 200 were classified as MFA by one of three or by Timo's manual review. The combined MFA cohort — everything Timo was, in the end, willing to describe as MFA — was 1,352 domains, or 33% of the delivered domain set.
The spend against that MFA cohort was £470,000 out of the total £1.24m spent in the year, or approximately 38%.
Two things about this number are worth naming.
The first is that Timo was neither shocked nor unshocked by it. He had, from a decade in the industry, expected an MFA share somewhere in the 25-45% range. His account fell in the middle of that range. The industry norm, on the brands we audit, is roughly consistent with this estimate.
The second is that the £470,000 had not, on inspection, produced no value. The MFA sites had delivered impressions, at cheap CPMs; some viewability was recorded (though the viewability signal on MFA inventory is well-documented to be inflated by the underlying site design). Some downstream conversion attribution had been credited to those impressions. Whether the value was real is a different question — but the account had, on the platform-reported view, "worked" in some sense against MFA delivery.
The reallocation
Timo's proposal, presented to his head of marketing three weeks after the audit was complete, was to implement systematic MFA exclusion across all campaigns, with an initial exclusion list of the 812 domains that had cleared all three third-party classifications, followed by a monthly review to add domains to the exclusion list as they surfaced.
The head of marketing's response, when Timo brought her the number, was — in Timo's words — "not what I had expected". She was less concerned about the current-state waste than she was about what excluding the MFA cohort would do to the reported delivery numbers on the account. Impressions would fall. Cheap CPMs would rise, because the account would now be competing exclusively against higher-quality inventory. The dashboard, in the short term, would look worse.
She was, on Timo's honest reflection, correct about all of these things. The head of marketing was under monthly delivery-report pressure from her CMO. Timo's proposal was going to make her monthly report harder to defend, at least in the first quarter after implementation, even though it would make the account's real-world performance meaningfully better.
The negotiation, in the end, took two months. The compromise they arrived at was to roll out the MFA exclusion over a three-month window rather than all at once, with a parallel expansion of the buying against curated marketplaces (which offered comparable delivery volumes at higher CPMs against genuinely reviewed inventory). The dashboard's short-term deterioration was smoothed by the parallel expansion. The CMO was briefed on the change in language that emphasised the medium-term improvement rather than the short-term impression-count decline.
"The technical work was straightforward. The internal politics was the hardest part. The MFA problem in the industry is, at least in part, sustained by the political difficulty of making the fix visible without having to explain why it wasn't fixed sooner."
What happened
Six months into the reallocation, the account's numbers looked meaningfully different.
Total impressions, on a like-for-like campaign basis, were down approximately 24%. This was, as predicted, uncomfortable to defend on the dashboard. Blended CPMs were up approximately 45%, from an average of £2.20 to an average of £3.20. Blended CTR was up meaningfully — from 0.14% to 0.31% — because the impressions now being counted were on inventory the audience was more likely to actually engage with.
The number Timo was most interested in was blended cost per meaningful downstream action. In the brand's case, this was primarily lead form submissions from the account's product-consideration pages. Blended CPA on those actions, on the reallocated account, fell approximately 34% quarter over quarter and continued falling through the second and third quarters after the reallocation. By month six, the account was producing roughly 41% more downstream leads per pound spent than it had been producing pre-reallocation.
The MMM partner the brand used ran an independent analysis six months in. Their attributed incremental ROAS on the programmatic line was up approximately 60% year-over-year, on a spend base that was flat. On the MMM number — which is the number the CFO cared about — the reallocation had produced a substantial improvement in the account's productivity.
The political aftermath
The head of marketing's position, when the six-month numbers came in, was defensible. She had made a decision, in Q1, to slow-roll the reallocation to protect the dashboard's short-term appearance. The decision had, in retrospect, cost the brand something — roughly £30-40k of avoidable spend, on Timo's estimate — but had also protected her from a difficult CMO conversation at a moment when the CMO was under pressure from other quarters. The CMO, who saw the six-month numbers, was pleased with the outcome and not particularly interested in interrogating the sequencing of how it had happened.
Timo has, since the audit, run a similar exercise on two other accounts he now consults on. Both produced MFA shares in the same 30-40% range. Both required similar internal negotiation to implement the fix. Neither took as long as the first, because Timo has since developed a briefing template that makes the political conversation more efficient. The template, in short, is: describe the problem in terms of the CFO's likely reaction if the current state persists into next year's finance review, rather than in terms of the immediate impression-count implications.
The technical fix is straightforward. The political framing is not. On the accounts we now watch, the political framing is what determines whether the fix actually gets deployed. Head of programmatic teams have, in general, been under-trained in the political framing. The successful ones — and Timo, on his three-for-three record, is among them — have been forced to learn it on the job.