The private marketplace — I have agreed not to name it, and the individual buyers who agreed to talk asked to remain anonymous — was, until very recently, marketed as a curated premium supply pool available exclusively to a small number of vetted programmatic buyers. Seven brands were buying against it, at CPMs ranging from £14 to £22 depending on the specific deal. The PMP's marketing described the inventory as "publisher-reviewed", "brand-safe", "not available on open exchange", and — most consequentially — "exclusive to our participating buyers".
The last claim was, on the log-level audit two of the seven brands eventually ran, false.
The audit
The audit began with a coincidence. One of the seven brands, running a broad-market open-exchange inclusion list on a separate DSP, noticed that a particular publisher-controlled ad-slot ID was appearing in both the PMP delivery reports and in the open-exchange delivery reports, in the same weeks, on the same publishers, at substantially different CPMs. The PMP was clearing at £16.40 average. The open exchange was clearing, against what appeared to be the same slot, at £6.20.
The head of programmatic at the brand — call him Björn — asked his DSP account team for an explanation. The DSP account team, after some internal consultation, said that the ad-slot IDs were, in fact, the same, but that the PMP inventory was "curated at the impression level" — meaning that individual impressions the PMP served were, in some way, higher-quality than individual impressions the open exchange served, even where both were sold against the same slot ID. Björn asked for the audience-verification data that would support this claim. The DSP could not produce it. The PMP, when asked directly, also could not.
Björn shared his finding with a peer at one of the other participating brands. That head of programmatic ran the same analysis on his own account and found the same pattern. Over the next six weeks, a small coalition of the participating brands ran a coordinated log-level review across their DSP delivery data, matching PMP-delivered impressions against open-exchange-delivered impressions on the same slots, in the same weeks, from the same publishers.
The finding, on the aggregate data, was unambiguous. Approximately 60% of the PMP's delivered impressions were, on any technical read, indistinguishable from impressions that the same publisher was simultaneously selling in the open exchange at approximately 40% of the PMP CPM. The remaining 40% of PMP impressions were on inventory that did not appear in open-exchange delivery — but this share was much smaller than the PMP's exclusivity claim had implied.
What the PMP said
The buyers, once they had characterised the finding, approached the PMP directly. The PMP's response, given in a meeting with the coalition's lead buyer, was — in Björn's word — "circular". The PMP maintained that the inventory sold through the PMP path was, at the impression level, higher-quality than the inventory sold through the open-exchange path, on the grounds that PMP delivery included additional quality-control measures the PMP applied internally. When asked to describe those measures, the PMP declined to be specific, on the grounds that the measures were commercially confidential.
The buyers, unsatisfied, asked their DSP to run the impression-level analysis independently. The DSP's own analysis broadly confirmed the buyers' finding: the majority of PMP-delivered impressions were coming from publisher inventory that was simultaneously available at meaningfully lower CPMs through the open exchange, with no reliable technical or audience-quality distinction between the two.
The PMP, when confronted with the DSP's independent analysis, revised its position. The new position was that the PMP's value proposition was not the exclusivity of the inventory but the "publisher partnership" the PMP maintained, which produced value at levels that impression-level analysis could not capture. This position was, in the buyers' polite reading, marketing language covering a substantive inability to justify the CPM differential.
"The PMP's argument, in the end, was that we should trust the value they added because they added value we could not measure. This is the same argument the open exchange rejected a decade ago when SSPs were pricing themselves above their measured performance. The PMP world has, on this evidence, not learned the lesson."
What the buyers did
Five of the seven brands ended their PMP arrangements within eight weeks of the finding. Two continued at renegotiated CPMs — approximately 35% below the previous rates — with explicit contractual language about the exclusivity of the inventory and the buyer's right to audit at the impression level.
The reallocated PMP spend, across the five exiting brands, went to a mixture of destinations. Roughly 40% went to open-exchange delivery against the same publishers, at the open-exchange CPMs. Roughly 30% went to genuinely-exclusive PMPs with other publishers (the buyers doubled the verification requirements on these arrangements and were, on aggregate, more skeptical of exclusivity claims than they had been previously). The remaining 30% went to broader open-exchange delivery against inventory-quality-filtered supply — the same supply-path optimisation approach we have written about elsewhere.
Blended CPMs on the reallocated spend, in aggregate, fell approximately 27% against the previous PMP-heavy allocation. Blended delivery volume on the same spend rose approximately 34%. Reported viewability was broadly flat. Attributed downstream performance, on the MMM analyses two of the brands ran independently, was up meaningfully — the buyers had not, in the end, been getting substantive extra value from the PMP that would have justified the CPM premium.
What this signals
The specific PMP in this story is not, on our reading, an outlier. In the twelve months since Björn's coalition ran the audit, we are aware of at least three other similar findings against different PMPs — one in the news category, one in health, one in a specific vertical publisher network. In each case, the impression-level analysis has revealed that the exclusivity claim on which the PMP's premium was priced does not, in aggregate, hold up.
The underlying dynamic is not particularly complicated. Publishers who sell inventory through multiple paths — direct deals, PMPs, open exchange — have a commercial incentive to sell the same inventory through the highest-CPM path that will clear it. Where the buyers on each path are not comparing notes, the publisher can, in effect, price-discriminate against the highest-paying path. This is legal, in most cases; it is defensible under most PMP contracts; and it is, on the evidence we now have, more widespread than the buyer community has fully appreciated.
The specific practical implication for programmatic buyers is that PMP arrangements should, henceforth, include explicit contractual language requiring the publisher to disclose whether the same inventory is being sold through other paths, at what CPMs, in what volumes. Where the publisher declines to include such language, the PMP should be priced against the assumption that the exclusivity is a marketing claim rather than an operational reality. This is a substantial shift in how PMPs have been priced for the last decade. It is, on the evidence, a shift buyers should have made sooner.