A UK competition authority measured it: publishers receive around 65% of what advertisers pay, so roughly 35 pence in every pound stops somewhere in between. That is the headline, and it is the most conservative of the published estimates.

This is not a conspiracy. Every layer of the chain performs a function somebody is willing to pay for. The problem is that almost no advertiser can name the layers, price them, or say what their own campaign paid.

This page walks the chain, gives what each measurement found, and says plainly where the numbers disagree and why.

What programmatic actually is

Strip the vocabulary and the mechanism is simple. The complexity is in who takes a cut on the way through.

The transaction. A page loads. An impression becomes available. Eligible advertisers are evaluated, one wins, and an ad renders. The whole thing resolves in the time the page takes to appear.

The chain, for anything not bought directly from a platform. The publisher offers inventory through a supply-side platform. The advertiser bids through a demand-side platform. The two meet on an exchange. A media agency may sit above the advertiser, and ad servers sit on both sides.

What each layer charges. A percentage of the transaction, a fee per thousand impressions, or both. None of it is visible on the advertiser’s invoice as a separate line unless the advertiser demanded it be.

Why the opacity is structural rather than deliberate. Each intermediary sees its own leg of the transaction. No single participant observes the whole path from advertiser payment to publisher receipt, which is why measuring the chain requires matching log files from a dozen companies who do not normally share them.

Why that matters before any figure below. Every measurement in this article exists because somebody organised that matching exercise. There are only a handful, they are hard, and their results do not fully agree.

What the regulator measured

The most careful public estimate comes from a competition authority that could compel the data, which is the only way this measurement gets made properly.

The headline. The UK Competition and Markets Authority, in its 2020 market study on 2019 data, estimated that “on average in 2019, publishers received around 65% of initial advertising revenue that was paid by advertisers (ie the overall ‘ad tech take’ was around 35 pence from every pound spent by advertisers).”

The regulator’s own gloss on that. “It is striking that collectively they are able to take more than a third of the total amount paid by advertisers.”

The layer-by-layer take rates it measured. Weighted averages across the platforms that supplied data, each expressed as a share of the spend passing through that layer: 14% for demand-side platforms, ranging from 5% to 42%; 16% for supply-side platforms, ranging from 8% to 25%; 34% for ad networks, ranging from 25% to 42%; and 5% for media agencies, ranging from 4% to 14%.

Buy-side total. “Our estimate is that buy-side fees are equivalent to 23% of overall advertising expenditure.”

What the regulator says about its own figure. That publishers’ share “may be lower than this average”, for four reasons it lists: trading-desk and third-party data fees not covered, an additional charge on one bidding product, categories another study could not attribute, and ad fraud.

The date, which matters. This is 2019 data published in July 2020. It predates several antitrust rulings and significant changes in how the chain operates. It remains the most rigorous public estimate, and it should always be quoted with its year.

The study that could not account for 15% of the money

A parallel industry study of the same market reached a lower figure, and its most useful finding is what it could not find.

Who ran it and on what. ISBA and the AOP commissioned PwC to match log files across “15 advertisers, 12 agencies, five DSPs, six SSPs and 12 publishers, representing approximately £0.1bn of UK programmatic ad spend”. Data collection ran 1 January to 20 March 2020.

What reached publishers. “Publisher revenues in our study ranged from 49% to 67% of advertiser spend for each individual publisher. When giving equal weight to each supply chain, the average was 51%.”

The finding that made the study famous. “15% of advertiser spend, the unknown delta, representing around one-third of supply chain costs, could not be attributed.” Across individual supply chains it ranged “from 0% to 86%, with the majority from 2% to 23%.”

What the study says the delta might be, in its own words. “This study cannot say with any certainty what the unknown delta represents. It could reflect a combination of: limitations in data sets, necessitating occasional estimations; DSP or SSP fees that aren’t visible in the study data; post-auction bid shading; post-auction financing arrangements or other trading deals; foreign exchange translations; inventory reselling between tech vendors; or other unknown factors.”

The number that should worry you more than the delta. “From 267 million impressions served from study advertisers to study publishers, 31 million (12%) were successfully matched. The rest could not be mapped due to low data quality.”

Read that twice. Fifteen advertisers and twelve publishers who had agreed to cooperate, with a Big Four firm running the reconciliation over fifteen months, could match twelve percent of the impressions. That is the state of traceability in the chain your budget passes through.

One update the figure deserves, and it points the right way. The US study cited below notes that “ISBA’s follow-up study in 2023 revealed just a 3 percent unknown delta due to improved data matching”, and reports finding no delta itself on the subset it could match one-to-one. Quote the 15% as a 2020 measurement rather than as the current state, because matching has improved since.

The study’s own warning about generalising upward. “It’s important to realise that this study represents the most premium parts of programmatic: the highest profile advertisers, publishers, agencies and adtech. If examined, the ‘long tail’ would presumably further reinforce these findings.” The 51% is a ceiling, not a market average.

Take rates by intermediary layer measured by a competition authorityChart presenting the take rates measured by a competition authority for each layer of the programmatic advertising supply chain, using 2019 data published in July 2020, each rate expressed as a weighted average share of the spend passing through that layer rather than of total advertiser spend. Demand-side platforms took a weighted average of fourteen percent, with the lowest average at five percent and the highest at forty-two percent. Supply-side platforms took a weighted average of sixteen percent, ranging from eight to twenty-five percent. Advertising networks took a weighted average of thirty-four percent, ranging from twenty-five to forty-two percent, the authority noting that their fees are generally higher than those of demand-side and supply-side platforms. Media agencies took a weighted average of five percent, with estimates ranging between four and fourteen percent of total open display expenditure. Publisher advertising servers charged low fees typically on a cost per thousand basis, assessed at approximately zero to five percent of total media spend. The authority separately estimated that buy-side fees taken together were equivalent to twenty-three percent of overall advertising expenditure, and that publishers received around sixty-five percent of the initial advertising revenue paid by advertisers, meaning an aggregate intermediary take of around thirty-five pence in every pound.What each layer takesWeighted average, as a share of spend passing through that layer. 2019 data.Ad networks34%range 25 to 42Supply-side platforms16%range 8 to 25Demand-side platforms14%range 5 to 42Media agencies5%range 4 to 14Buy-side fees together: 23% of overall advertising expenditure.Publishers receive around 65%, so the aggregate take is around 35 pence in the pound.The regulator adds that publishers’ share “may be lower than this average”, and lists four reasons.
Weighted averages from platforms compelled to supply data, 2019. The ranges matter more than the averages. Source : Competition and Markets Authority (2020)
Impression matching rate and unattributed spend in a programmatic supply chain studyChart presenting the traceability findings of a programmatic supply chain transparency study conducted over the first quarter of two thousand and twenty. The study collected data from fifteen advertisers, twelve agencies, five demand-side platforms, six supply-side platforms and twelve publishers, representing approximately one hundred million pounds of United Kingdom programmatic advertising spend, and reviewed two point two billion lines of data across two hundred and ninety unique supply chains over fifteen months. Of two hundred and sixty-seven million impressions served from study advertisers to study publishers, only thirty-one million, or twelve percent, were successfully matched end to end, the remainder being impossible to map due to low data quality. Publisher revenues ranged from forty-nine to sixty-seven percent of advertiser spend for each individual publisher, averaging fifty-one percent when equal weight was given to each supply chain. Fifteen percent of advertiser spend, described as the unknown delta and representing around one third of supply chain costs, could not be attributed at all, ranging from zero to eighty-six percent across individual supply chains with the majority falling between two and twenty-three percent. The study explicitly declined to explain that delta, listing possible causes including limitations in data sets, platform fees not visible in the study data, post-auction bid shading, post-auction financing arrangements, foreign exchange translations and inventory reselling between vendors.What cooperation and fifteen months boughtImpressions observed267 millionImpressions matched31 million, or 12%Reached the publisher51%on average, 49% to 67%by individual publisherCould not be attributed15%on average, 0% to 86%across individual supply chainsThe study declined to say what the delta was.Its own list of candidates ran from data quality to bid shading to inventory reselling between vendors.
The unknown delta averaged 15% of advertiser spend and ranged from 0% to 86%. The study declined to say what it was. Source : ISBA and PwC (2020)

The US study, and the number that is not the same number

An American study of the same problem produced a figure that circulates widely and is routinely misread. Getting it right matters, because it measures something else.

The study. Twenty-one marketers and twelve supply chain companies, “$123 million in ad spend” and “35.5 billion impressions”, over a period running “between September 2022 and January 2023”.

What reached the seller. On an initial spend of $1,000, “$710 reaches the seller as revenue”, so 71%. That is the figure comparable to the 65% and 51% above, and it is the highest of the three.

The famous number, and what it actually counts. “After accounting for both transaction costs and loss of media productivity costs, only 36 cents of every ad dollar that enters a DSP effectively reaches the consumer.”

Why 36 cents is not a fee figure. It splits into “transaction costs, primarily DSP and SSP costs, accounting for 29 percent of the original ad dollar” and “loss of media productivity costs, including non-viewable and IVT impressions as well as non-measurable and MFA ad spend, accounting for 35 percent”. The second half is not money taken by intermediaries. It is money spent on impressions no human usefully saw.

The distinction to hold onto. 71% reaches the seller. 36 cents buys an impression that is viewable, valid, measurable and on a real site. Anyone quoting “only 36 cents reaches the publisher” has merged two different measurements.

And what the study left out. Agency fees and brand safety costs were outside its scope. Its own note: “Depending upon those costs, less than 36 cents of every dollar would effectively reach the consumer.”

Made-for-advertising sites, and the number of sites you are on

This is the finding with the most direct operational consequence, and it is the one most advertisers can act on this week.

What these sites are. Sites built to carry advertising rather than to be read. The study describes them as usually featuring “low-quality content, such as fake news, conspiracy theories, or spammy links”.

How much of the money lands there. “21 percent of impressions and 15 percent of spend” in the study. Individual advertisers ranged “from 0.13 percent to 42 percent”, and “every advertiser recorded at least some media spending on MFA websites”.

The dispersion figure behind it. “The average number of websites (and apps) for the 21 marketer study participants was 44,000”, with a range “between 3,627 and 222,534”. The study calls that “deeply concerning” and notes that “advertisers can reach a high percentage of target audiences using a few hundred websites”.

Read those two together. You are on forty-four thousand sites. You chose none of them by name. A fifth of your impressions land on properties that exist to serve ads. Both facts have the same cause: a buying model where you specify an audience and the chain decides the inventory.

What to do about it, concretely. Ask your buyer for the full domain report, sorted by spend. Not the top twenty. All of it. Then ask what an inclusion list of a few hundred named domains would cost in reach. That conversation reveals more about a programmatic setup than any dashboard.

Why this matters more in B2B than elsewhere. Your addressable market is a few hundred to a few thousand companies. Reach is not your constraint, and the argument for accepting forty-four thousand unnamed domains, that you need the scale, does not apply to you.

Number of websites per campaign and share of spend on made-for-advertising sitesDiagram presenting two findings from a United States programmatic transparency study covering twenty-one marketers, one hundred and twenty-three million dollars of advertising spend and thirty-five and a half billion impressions between September two thousand and twenty-two and January two thousand and twenty-three. The first finding is dispersion: the average number of websites and applications across the twenty-one participating marketers was forty-four thousand top-level domains, ranging from three thousand six hundred and twenty-seven to two hundred and twenty-two thousand five hundred and thirty-four, a figure the study describes as deeply concerning while noting that advertisers can reach a high percentage of target audiences using a few hundred websites. The second finding concerns made-for-advertising sites, meaning properties built to carry advertising rather than to be read, usually featuring low-quality content such as fake news, conspiracy theories or spam links: these accounted for twenty-one percent of study impressions and fifteen percent of spend, with individual advertisers ranging from zero point one three percent to forty-two percent of their spend landing there, and every advertiser in the study recording at least some. Both findings share a cause, namely a buying model in which the advertiser specifies an audience and the supply chain selects the inventory, and both matter more in business markets where the addressable universe is a few hundred to a few thousand companies and reach is not the binding constraint.Where the impressions actually landedWebsites per campaign44,000on average, from 3,627 to 222,534The study notes a few hundred sites wouldreach most of the target audience.Made-for-advertising sites21%of impressions15%of spendFrom 0.13% to 42% by advertiser.Every advertiser had some.Ask for the full domain report by spend. Not the top twenty. All of it.Then ask what an inclusion list of two hundred named domains would cost you in reach.That conversation reveals more than any dashboard.
You chose none of them by name. Both facts have the same cause: you specify an audience and the chain picks the inventory. Source : ANA Programmatic Media Supply Chain Transparency Study (2023)
Comparison of the published measurements of what reaches a publisher in programmatic advertisingTable comparing three published measurements of what share of advertiser spend reaches the publisher in programmatic advertising, alongside a fourth widely quoted figure that measures something different. A competition authority market study using two thousand and nineteen data estimated that publishers receive around sixty-five percent of initial advertising revenue, so an aggregate intermediary take of around thirty-five pence in every pound. An industry study covering fifteen advertisers and twelve publishers over the first quarter of two thousand and twenty found publishers receiving fifty-one percent on average, with individual publishers ranging from forty-nine to sixty-seven percent, and could not attribute fifteen percent of advertiser spend at all. A United States study covering twenty-one marketers, one hundred and twenty-three million dollars of spend and thirty-five and a half billion impressions between September two thousand and twenty-two and January two thousand and twenty-three found seven hundred and ten dollars of every thousand reaching the seller, so seventy-one percent. The fourth figure, thirty-six cents of every dollar, comes from that same United States study but counts something different: it is the share of a dollar that ends up buying an impression that is viewable, valid, measurable and on a genuine site, splitting into twenty-nine percent of transaction costs and thirty-five percent of lost media productivity, the latter not being money taken by intermediaries but money spent on impressions no human usefully saw. Anyone quoting thirty-six cents as the publisher’s share has merged two different measurements.Four numbers, and one of them measures something elseSTUDYFIGUREWHAT IT MEASURESUS study, 2022-202371%Reaches the seller$710 of every $1,000UK regulator, 2019 data65%Reaches the publisher”35 pence from every pound” takenUK industry study, 202051%Reaches the publisherPlus 15% that could not be attributed at allThe famous one36 centsNOT the publisher’s shareBuys a viewable, valid, measurable impression on a real site29% is transaction cost. 35% is impressions no human usefully saw.The second half is not money taken by intermediaries. Merging the two overstates the fees.
71%, 65% and 51% measure the publisher's share. The 36 cents figure measures something different and is routinely merged with them. Source : MASTRATOS from CMA, ISBA/PwC and ANA (2026)

What a court established about one take rate

Studies can be argued with. A judgment after a trial, with internal documents in evidence, is a different kind of source.

The finding. In its April 2025 memorandum opinion, a US federal court found that one exchange “maintained an overall take rate of around 20%” and described “AdX’s charging a durable 20% take rate for well over a decade” as “direct evidence that Google has possessed monopoly power in the open-web display ad exchange market.”

What competitors charged. The same opinion records that “Google employees recognized that AdX’s 20% take rate was higher than that of rival ad exchanges like IndexExchange, Magnite, and Xandr, which often charged closer to 10%.”

How the rate varied without varying. “AdX would charge a 15% or lower take rate for competitive impressions, and a 25% or higher take rate for less competitive impressions, all the while maintaining an average take rate of around 20%.” Your individual impression was priced by how much competition it faced, and the average concealed that.

The negotiating position it implies. The court noted evidence that “only six of AdX’s 3,815 largest publisher customers received discounts.”

The layer that costs almost nothing, for contrast. Publisher ad server fees “typically average about 1% to 2% of the revenue generated by the sale of the impression”, and Google’s own defence expert put its ad server take rate at “1.7 percent to 1.3 percent”.

One figure from the same opinion worth using carefully. “Typically, between $20 to $40 of every $100 spent on digital advertising goes to the ad tech companies that develop and operate these tools.” That is the whole industry collectively, not one company. It is also the range within which every study above falls.

A number that is not in the opinion. The widely repeated claim that Google keeps around 35 cents of every advertiser dollar across its stack does not appear in that judgment. It comes from the complaint. If you quote it, quote it as an allegation.

Exchange take rate established in a federal court judgmentChart presenting findings on advertising exchange take rates established in a United States federal court memorandum opinion issued in April two thousand and twenty-five. The court found that one exchange maintained an overall take rate of around twenty percent, describing its charging of a durable twenty percent take rate for well over a decade as direct evidence of monopoly power in the open-web display advertising exchange market. The same opinion records that employees of the operator recognised this rate as higher than that of rival exchanges, which often charged closer to ten percent. The court also described how the rate was modulated without the average changing: the exchange would charge fifteen percent or lower on competitive impressions that had received relatively high offers from third-party exchanges, and twenty-five percent or higher on less competitive impressions, all the while maintaining an average of around twenty percent, so an individual impression was priced by how much competition it faced while the average concealed that variation. On negotiating leverage, the opinion notes evidence that only six of the exchange’s three thousand eight hundred and fifteen largest publisher customers received discounts. For contrast, publisher advertising server fees typically average about one to two percent of the revenue generated by the sale of an impression, and the operator’s own defence expert put its advertising server take rate at between one point three and one point seven percent.What a trial established about one layerOne exchange20%“durable”, over a decadeRival exchanges~10%“often charged closer to”Publisher ad server1 to 2%a different order entirelyThe average concealed the variation15% or lower on contested impressions,25% or higher where competition was thin.Negotiating leverage6 of 3,815large publishers got a discount
Varied between 15% and 25% by how contested the impression was, while the average held at 20%. Six of 3,815 large publishers got a discount. Source : US District Court, Eastern District of Virginia (2025)

What this means if you are a B2B advertiser

Everything above is context. Here is what changes on your account.

The chain costs roughly a third before anyone sees anything. Whatever your view of whether that is fair, budget for it. A programmatic display line item is not comparable to a search line item at the same nominal spend.

Cheap inventory is cheap for a reason. The impressions available at low prices are the ones nobody with better options wanted. In a market of a few hundred addressable companies, buying more of that is not a strategy.

Retargeting is the defensible use. A known audience, a short list of domains, a measurable outcome. This is where programmatic earns its keep in B2B, and it is a small budget line rather than a channel strategy.

Broad programmatic awareness rarely survives the arithmetic. You are paying a third to intermediaries to reach forty-four thousand unnamed sites on behalf of a target of a few hundred companies. The reach argument that justifies this for consumer brands does not transfer.

Three questions to put to whoever runs it. Show me the full domain report by spend, not the top twenty. What percentage of our spend landed on sites we could not name in advance? What would an inclusion list of two hundred named domains cost us in reach?

The answer that should end the conversation. If nobody can produce the domain report, you are not buying media. You are buying a promise about media, and the studies above describe what happens to money bought that way.

How to read a programmatic proposal

Most proposals describe capabilities. The useful reading looks for what is absent.

Look for the fee disclosure first. Is there a line separating media spend from platform fees and from the agency’s own margin? If everything is one number labelled “budget”, nothing downstream can be computed, and the studies above tell you roughly what that opacity is worth.

Ask which platforms are in the path. A demand-side platform, an exchange, a supply-side platform, possibly a data provider and a verification vendor. Each has a take rate. A proposal that names none of them is describing an outcome, not a supply chain.

Ask whether the buying is on an inclusion list or an exclusion list. An exclusion list means you buy everything except what you blocked, which is how a campaign ends up on forty-four thousand domains. An inclusion list means you buy only what you named. The second costs reach and buys certainty, and in B2B that is usually the right trade.

Ask what happens to the data. Whether you keep the domain-level logs, whether they are available after the campaign ends, and whether they come from the platform or from the agency’s own reporting layer. The difference matters when you want to check anything.

Ask who owns the seat. If the buying runs through the agency’s own platform account rather than yours, your campaign history, your audience lists and your negotiated rates leave with the agency. This is invisible until the day you change supplier.

The proposal you want is shorter than the one you will get. Named platforms, separated fees, an inclusion list, log access, and your own seat. Everything else is presentation.

Where to go next

You want the whole chain and its vocabulary. Media buying explained.

You are buying exposure and want to price it. CPM, cost per thousand impressions.

You want search rather than display. Google Ads account structure for B2B.

You are splitting a budget between platforms. How to split budget between Google and Meta in B2B.

Your counters disagree with each other. Why GA4 and Meta conversions don’t match.

You want to know which metric to run on. ROAS vs MER vs CAC vs LTV.

In short

  • A regulator measured publishers receiving around 65% of advertiser spend on 2019 data, so an aggregate intermediary take of about 35 pence in the pound.
  • An industry study of the same market found 51%, and could not attribute 15% of spend at all, ranging from 0% to 86% across supply chains.
  • That study matched 12% of the impressions it observed. Fifteen cooperating advertisers, twelve publishers, fifteen months, and 88% could not be mapped.
  • A US study found 71% reaching the seller but only 36 cents of every dollar buying a viewable, valid, measurable impression on a real site. Those are two different measurements and they are routinely merged.
  • Made-for-advertising sites took 21% of impressions and 15% of spend, and every advertiser in that study had some.
  • The average campaign ran across 44,000 websites, from 3,627 to 222,534, where “a few hundred websites” would reach most of the target.
  • A federal court found a durable 20% take rate at one exchange for over a decade, against rivals “closer to 10%”, varied between 15% and 25% by how contested the impression was.
  • In B2B, retargeting is the defensible use. Broad programmatic awareness rarely survives the arithmetic on a market of a few hundred companies.

If nobody can show you the domain report, you are buying a promise about media. Book a diagnostic, or see how we approach B2B paid acquisition.