What win CPM actually measures
Win CPM is the average price a buyer actually pays per 1,000 impressions on the auctions it wins. A demand-side platform records the clearing price from each auction it wins and averages those prices. That matters because your bid and your clearing price are not always the same thing. In a first-price auction they are identical. In a second-price auction you can pay the runner-up bid plus one cent, or the floor when that is higher.
OpenRTB carries that clearing price back through win notices and billing notices. The AUCTION_PRICE macro reports the final price after any seller discount. Win CPM built on win notices can diverge from one built on billing notices because a win does not always mean billable delivery. Billing notices are where actual spend is applied.
Why first-price changed what the number reveals
For most of the 2010s, programmatic display was largely second-price. You could bid your true value and let the auction discount you. The shift to first-price between 2017 and 2019 changed that. Google announced unified first-price auctions in March 2019 and began rolling out in September. In a Q2 2018 test, Hearts & Science measured CPMs 59% higher under first-price than second-price, and still 54% higher with bid shading.
Bid shading became the buy-side answer. DSPs train models on historical clearing data to predict the minimum winning price and bid just above it. Adobe describes passing every proposed bid through a clearing price prediction model after pacing logic. Win CPM now largely reflects your own shading model as much as market conditions.
What the number hides
A low win CPM is not automatically a win. Brian O’Kelley’s example shows the trap: one DSP spends $100 at a $2.50 CPM and wins 3.1% of auctions; another spends the same at $1.10 CPM and wins 0.6%. The second sees more inventory and picks cheaper impressions. A low win CPM can simply mean you are only winning cheap, uncontested auctions.
Fees are another hidden layer. OpenRTB clearing prices are usually net of fees, while an advertiser’s invoice may include platform fees. Identical $5 bids can deliver between $2.40 and $3.83 to a publisher depending on fee structures and supply paths. Comparing win CPM across DSPs can mean comparing different definitions.
Adjacent terms worth knowing:
- Bid CPM: the average price offered, including losing bids. Under first price it converges with win CPM on won impressions.
- Minimum bid to win: the lowest price that would have secured a given auction, returned as AUCTION_MIN_TO_WIN.
- eCPM: publisher revenue normalised to a per-thousand figure, not a buyer clearing price.
- Floor price: the seller’s minimum, which can shape win CPM when only one bid clears.
What to do with it in 2026
Win CPM should sit alongside win rate and minimum bid to win in your reporting. If win CPM drops while win rate crashes, you are not getting cheaper inventory; you are losing auctions and seeing a narrow slice. If win CPM equals your bid almost every time, that first-price signal tells you to refine your shading or test lower bids.
Recent market data adds context. DataBeat’s June 2026 report found conventional programmatic buyers clearing at $6.95 CPM versus $6.13 for agentic buyers, a 13.4% gap, while agentic demand entered 86% fewer auctions. Its July report showed US programmatic CPMs up 51% year over year, with web at $1.42 and app at $1.70 compared with $1.13 a year earlier.
Transparency disputes and court remedies also affect comparisons. The FTC and 22 states allege Amazon placed an undisclosed soft reserve on search ads, with 79.1% of Sponsored Products clicks charged at the advertiser’s own bid in 2024; Amazon disputes the case. Judge Leonie Brinkema’s September 2, 2026 remedies include per-bidder floors, which can mean two buyers face different minimums on one impression, making cross-DSP win CPM comparisons harder to read.
The practical rule: treat win CPM as a starting point, not a scoreboard. Pair it with win rate, minimum bid to win, billable delivery and your own fee-adjusted cost per acquisition. The average price you paid is only useful if you know what you gave up to get it.
Source: PPC Land
