A Dollar of Display Advertising Reaches the Publisher as Fifty Cents

The Journey of One Impression

A reader opens a page. Before it finishes rendering, an auction has run and finished. The sequence looks roughly like this.

The publisher ad server notices an empty slot and passes it to a supply side platform, software representing the seller. The SSP broadcasts a bid request describing the slot and whatever is known about the reader to an ad exchange. On the other side, demand side platforms representing advertisers evaluate the request against campaign rules, consult data providers about the audience, and return bids. The exchange runs the auction, the winning creative loads, and a verification vendor later confirms the ad was actually viewable and shown to a human.

That entire round trip is real time bidding, and it completes in roughly one hundred milliseconds, billions of times a day.

Every Step Has a Rate Card

Each participant takes a cut, usually as a percentage of media spend rather than a fixed fee, which means the tolls compound.

Studies commissioned by advertiser trade bodies have repeatedly found that roughly half of a programmatic dollar reaches the publisher, with the rest absorbed by the chain. The precise figure varies by study, market, and deal type, and defenders correctly note that these intermediaries perform real functions. The point is not that the take is illegitimate. It is that most advertisers had no idea it was that large.

Why Reasonable Fees Produce Half a Dollar

The word compound in that opening sentence is carrying the whole explanation, and working it through shows that no villain is required to reach the fifty percent figure. Illustrative and round.

Each layer takes a percentage of what reaches it, not of the original budget. So the money passes through a series of multiplications rather than a single subtraction.

Suppose four layers each take fifteen percent. Fifteen percent for software that bids across billions of requests a day. Fifteen percent for running the auction. Fifteen percent for representing the publisher. Fifteen percent for planning and managing the campaign.

Argue with any one of those in isolation and you have a difficult case. Each is a real service with real engineering behind it, and fifteen percent is not an outrageous price for any of them.

Now put them in series. What survives is eighty five percent of eighty five percent, four times over, which is about 52.2 percent. The publisher receives a little over half, and every participant charged a defensible rate.

That is the finding those industry studies keep reproducing, and it arrives without anybody overcharging. It is what a chain of percentage fees does.

The consequence for anyone hoping to fix it is bleak. Suppose every layer halved its take to around eight and a half percent, which no intermediary would accept quietly. The publisher's share rises to roughly seventy percent. An enormous concession across the entire industry buys about eighteen points.

So squeezing fees is not the lever. The number of layers is the lever, because each additional one multiplies rather than adds, and removing a layer entirely is worth far more than negotiating all of them down.

The Part Nobody Could Account For

The finding that genuinely unsettled the industry came from an attempt to reconcile spend end to end. A widely cited 2020 study by a UK advertiser association, working with a major accounting firm, tried to match advertiser payments to publisher receipts impression by impression across a set of real campaigns.

It could only match a minority of impressions at all, and it identified roughly fifteen percent of spend as an unknown delta: money that left the advertiser, did not arrive at the publisher, and could not be attributed to any identified intermediary. Not fraud necessarily, and not proven theft, but genuinely untraceable within the industry own record keeping.

The Impression That Was Never Seen

Underneath the question of where the money went sits a harder one about whether the thing being bought existed at all, and it is what the verification layer in the opening sequence is there to check.

Two failure modes attack the product itself.

The first is viewability. An advertisement can be requested, won at auction, loaded, counted, and billed while sitting far below the fold on a page nobody scrolled, or in a browser tab opened and never looked at. Every system in the chain records a successful delivery. No human eye was involved at any point. The impression was served and it was not seen, and those are different things that the plumbing has trouble distinguishing.

The second is non human traffic. Automated systems can generate page loads, and page loads are what the money follows, so there is a direct and obvious financial incentive to manufacture them. Fabricating traffic is cheap and the chain pays out on volume.

Notice the shape of the response, because it repeats the pattern above. The industry did not remove the participants selling unseen or fabricated impressions. It added verification vendors to measure how many of them there were.

That layer charges per impression, which means the buyer pays to have every impression checked, including all the legitimate ones. Measuring the problem became a permanent cost applied to the whole spend rather than to the part that was defective.

And it leaves the buyer in a strange position. They are paying one intermediary to confirm that the other intermediaries delivered what they charged for, in a market where they cannot see the underlying records themselves. In an industry where the buyer could reconcile a payment against a delivery, none of that apparatus would need to exist.

Header Bidding Was the Publishers Counterattack

Publishers noticed early that the auction was structured against them. Legacy systems called exchanges in sequence, in a waterfall, giving the first exchange in line an advantage that had nothing to do with price. Header bidding was the workaround: code placed in the page header solicits bids from many exchanges simultaneously before calling the primary ad server, forcing a genuine simultaneous auction.

It raised publisher yield materially and became standard. It also added latency to page loads and another layer of technical complexity, which is a recurring pattern here. Nearly every fix for a structural problem in this chain has been an additional layer rather than a removal of one.

Why Every Fix Adds a Layer

That closing observation is worth pursuing, because the pattern is not a coincidence and it explains why the chain keeps getting longer.

Consider what any participant in this market can actually do. It controls its own software and its own contracts. It does not control the ad server, or the exchange, or the verification vendor, because those belong to other companies with their own customers and their own revenue.

So when a participant identifies a problem, the only remedy available is to build something and insert it. Nobody in the chain has the power to delete a link in it.

The record follows exactly that shape. Publishers disadvantaged by the waterfall did not remove the waterfall, they added header bidding in front of it. Advertisers worried about fraud and viewability did not remove the sellers of fraudulent inventory, they added verification vendors to measure it. Buyers wanting better targeting added data providers.

Each fix worked. Each also arrived with its own percentage or per impression fee, which is subtracted from the same dollar, so part of the gain from the remedy is consumed by the remedy.

Header bidding is the clean illustration. It genuinely raised publisher yield by replacing a rigged sequence with a real auction. It also added latency, engineering burden, and more participants taking a share of the same impression. The publisher ended up better off and the chain ended up longer.

Which is why the complexity only ever increases. In a market where every actor can add and none can subtract, the number of layers is a ratchet, and the previous section showed exactly what a ratchet on layers does to the publisher's share.

When the Auctioneer Owns Both Sides

The structural conflict at the center of the industry is that the largest participant operates the publisher ad server, a leading exchange, and a leading buying tool at the same time. That is a party running the auction while also representing both the buyer and the seller.

Antitrust authorities pursued this directly, and in 2025 a federal court found unlawful monopolization in publisher ad serving and ad exchange markets, along with unlawful tying between them, with remedies litigated separately. However the remedy lands, the ruling established that owning multiple non adjacent positions in one auction chain is a legal exposure and not merely a governance concern.

Why That Position Cannot Be Audited From Outside

It is worth being precise about why holding all three roles is troubling, because the objection is structural rather than an accusation about any particular behaviour.

Think about what each position confers. The ad server decides which impressions are offered and on what terms, which is control over what gets auctioned at all. The exchange sets the rules of the auction and sees every bid submitted. The buying tool participates in that auction on behalf of advertisers.

Hold all three and you are writing the rules, observing everyone's hand, and playing. Even assuming impeccable conduct throughout, the counterparties have no way to confirm it, because the only complete record of what happened belongs to the party whose conduct is in question.

That is where this connects back to the unreconcilable fifteen percent. A buyer who cannot match payments to deliveries cannot audit the auction either, and the reason is the same in both cases: the log is not theirs. The opacity and the conflict are not two separate problems, they are one problem seen from two angles, since the conflict is only tolerable if it can be verified and the opacity is exactly what prevents verification.

Which is the real significance of the 2025 finding. It moved the question from a governance debate, where a company can respond with assurances and internal policies, to a legal one, where structure itself is the exposure and the remedy is decided by somebody other than the incumbent.

Why the Chain Survives Anyway

Given the leakage, the obvious question is why buyers and sellers do not simply transact directly. Some do, through programmatic guaranteed and private marketplace deals, and that share has grown. But direct deals require sales teams, negotiation, and reconciliation for every counterparty, and the open auction offers something they cannot: instant access to essentially all inventory with campaign level targeting and pacing.

The chain persists because it sells reach and automation, and because the cost of it is expressed as a percentage buried in a media plan rather than as an invoice line. Costs that never appear as a bill are the hardest ones for any organization to attack.

The Bottom Line

Programmatic advertising is an extraordinary piece of engineering wrapped around an unusually opaque commercial structure. Roughly half the money is consumed in transit, a slice of it cannot be traced at all, and the largest operator was found to be running an auction it also participated in on both sides. For anyone analyzing a media business the useful instinct is to ask not what the advertising market is worth, but which layer of it captures the money, because those are very different questions with very different answers. And the half that disappears is produced by the number of tolls rather than the size of any one of them, which is why negotiating rates has never fixed it.