Put a dollar into a demand-side platform. Picture it as a coin dropped into the top of a machine. It clatters through DSP fees, then a data fee, then an exchange, then a supply-side platform, maybe a second one, past a verification vendor, and out the bottom toward a publisher. By the time it lands on something a real person can actually see, the dollar is small change.
This is not a hunch. It is the headline finding of the most thorough audit the industry has ever run on itself. The Association of National Advertisers spent more than two years tracing log-level data on real campaigns, and concluded that only 36 cents of every ad dollar entering a DSP reached the consumer as a quality impression. The other 64 cents went somewhere. This post is about where.
Origin: how the chain got so long
For the first decade of web advertising, the path from advertiser to publisher was short. A buyer talked to a seller, or to an ad network in the middle, and a price was set. Then the impression became an auctioned item, and the middle filled up.
Each link was added to fix a real problem. The DSP appeared so a buyer could bid across many exchanges from one console. The SSP appeared so a publisher could manage which auctions its inventory entered. The data management platform appeared to attach audiences to the cookie. Verification firms appeared because nobody trusted the impression counts. None of this was designed as a single system. It accreted, layer on layer, and each layer charges.
Two later developments made the chain genuinely murky rather than merely long. The first was header bidding, the publisher technique that asks many demand sources for a bid at once. It killed the old ranked waterfall, which was good, but it also meant a single impression could now be offered down several routes at the same time, each with its own set of intermediaries and fees. The second was reselling. A supply-side platform that cannot reach a publisher directly can buy the same inventory from another SSP that can, then resell it. The impression arrives at the exchange two or three hops deep, and every hop took a cut. The buyer sees one line on an invoice. Behind it sits a chain nobody fully mapped. If you want the layout of who sits where, the ad-tech stack explainer lays out every box.
Present: the audit numbers, in order
The clearest picture comes from two bodies of work. In the UK, ISBA and PwC ran a forensic supply-chain study. In the US, the ANA did the same at larger scale. They used different samples and different years, and they agree on the shape of the problem.
Start with the ISBA waterfall, published in 2020. PwC pulled data from 15 advertisers, 12 agencies, five DSPs, six SSPs and 12 publishers, then tried to match impressions end to end. Of 267 million impressions observed, only 31 million, around 12 percent, could be matched cleanly, which is itself a finding about how opaque the chain is. For the impressions they could trace, the dollar split roughly like this: the agency fee took 7 percent, the DSP fee 8 percent, demand-side tech fees such as ad serving and verification 10 percent, the SSP fee 8 percent, supply-side tech 1 percent. Publishers received 51 percent. That left a gap.
The gap was 15 percent of advertiser spend that could not be attributed to anything at all. PwC named it the "unknown delta" and was admirably honest: the study could not say with certainty what it was. It could be data-set limitations, fees invisible in the data, post-auction bid shading, financing arrangements, or inventory reselling between vendors. The delta ranged from 0 to 86 percent across individual supply chains. A sixth of the money, in a "disclosed" model, simply could not be accounted for.
That 15 percent figure traveled. It became the number people quoted for everything wrong with programmatic. The more useful follow-up gets quoted far less. When ISBA and PwC ran the study again and published in 2023, the unknown delta had dropped to 3 percent, and the impression match rate had jumped from 12 percent to 58 percent. The delta was never mostly stolen money. It was mostly missing data, and better matching made it shrink. The ANA reached the same conclusion: with strict impression-to-impression matching of log-level data, it found a delta of zero. Hold that thought, because it draws the line between tax and waste.
The ANA study went further than ISBA in one direction: it asked not just whether the dollar reached a publisher, but whether it reached a quality impression. Its cost waterfall, built from $123 million in spend across 21 marketers and 35.5 billion impressions, splits the loss into two buckets.
The first bucket is transaction costs, the legitimate fees. On a $1,000 buy the ANA traced $80 to DSP platform fees, $20 to DSP add-on tools, $60 to DSP data fees, and $130 to the SSP. That is $290, or 29 percent, gone to the machinery. Note what is missing: agency and trading-desk fees were outside the study's scope, so the real take is higher than 29 percent for anyone buying through an agency. The ISBA study, which did count the agency fee, put it at 7 percent on its own.
The second bucket is loss of media productivity, and this is the part that stings. Of the $710 that did reach a seller, the ANA marked $95 lost to non-viewable impressions, $5 to invalid traffic, $150 to inventory that could not be measured for viewability at all, and $100 to made-for-advertising sites. Add it up and 35 percent of the original dollar bought impressions that were fraudulent, unseeable, unmeasurable, or close to worthless. What is left, the TrueAdSpend, is 36 cents.
Tax versus waste
Here is the distinction that should change how you read all of this. The 29 percent in transaction costs is a tax. The 35 percent in productivity loss is waste. They are not the same problem and they do not have the same fix.
The tax buys something. A DSP runs the bidding logic, the frequency capping, the optimization. An SSP runs yield management for the publisher. A verification vendor checks whether an ad was viewable and whether the traffic was human. Those are real services, and a market with zero fees would be a market with no infrastructure. The honest goal is not to drive the tax to zero. It is to know what the rate is, to confirm the service is worth it, and to stop paying for hops that add a fee and nothing else. A reselling chain that passes an impression through a third SSP is pure tax with no service attached.
Waste is different. A made-for-advertising site is built to harvest ad budgets, not to be read. It runs many ad slots per page, refreshes them aggressively, and buys cheap traffic from content-recommendation widgets. It games the metrics that are easy to game. MFA sites post strong viewability and low fraud scores, and their CPMs run about 25 percent below non-MFA inventory, which is exactly what makes them catnip for a DSP algorithm chasing cheap reach. Then they fail on the metric that matters: ads on MFA sites are at least 50 percent less likely to be linked to a sale than the internet average. The ANA found 15 percent of spend going there. Independent audits put real numbers on individual advertisers; Adalytics has documented brands that could be spending as much as $40 million a year on this inventory. Invalid traffic is the same story without the pretense: a bot, not a person. None of this buys anything. It is leakage, and the fix is to cut it out, not to negotiate it down.
The ANA's own framing makes the point. It estimated $22 billion in efficiency gains available in the open web, roughly one dollar in four, and most of that opportunity sits in the waste bucket, not the fee bucket.
Future and impact: the numbers are moving
The story is not static, and recent data is genuinely better. The ANA turned its one-off study into a quarterly benchmark with TAG TrustNet and Fiducia. By the Q1 2025 benchmark, 41 cents of the dollar reached the consumer, up from 36. Transaction costs had fallen to 26.1 percent. The biggest single mover was MFA spend, which collapsed from 15 percent in 2023 to around 1 percent, then lower still. By Q3 2025 transaction costs were down to 23.8 percent as SSP fees compressed. The pressure works. Buyers asking hard questions changed the rate.
The numbers are not all good news, and the honest reading matters. The ANA still measured $26.8 billion in unrealized media value in its Q2 2025 benchmark. A large share of spend still lands on inventory that cannot be measured for viewability, especially as connected TV grows. And independent investigators keep finding ads on MFA inventory even after the ad-tech firms that pledged to block it said they would. The gap between 41 cents and a dollar is still a canyon.
So what can a buyer who spends real money actually do. The audits converge on a short list. Know your take rate, line by line, which means insisting on log-level data; the ANA's whole method depends on it, and you cannot manage a fee you cannot see. Shorten the path. Every redundant SSP and every reselling hop is fee without service, and the fix has a name: supply-path optimization, the practice of buying the most direct route to an impression, covered in its own piece. Use inclusion lists, not endless exclusion lists, because the ANA found the average campaign spraying across 44,000 websites when a few thousand would do. And lean on curation, where a buyer or a trusted partner pre-assembles a marketplace of vetted publishers and strips out the resellers and MFA inventory before the auction runs.
The ad-tech tax is not a scandal to be abolished. It is a rate to be known and a path to be shortened. The advertisers moving the 36-cent number are not the ones who found a way to pay no fees. They are the ones who looked.
Council summary
This post argues that the missing two thirds of a programmatic dollar is not one problem but two, and that conflating them is why the waste persists. Anchored on the ANA and ISBA forensic audits, it separates the ad-tech tax, the roughly 29 percent in real DSP, SSP, and verification fees, from the roughly 35 percent in pure waste: fraud, unviewable and unmeasurable inventory, and made-for-advertising sites that the audits show are at least 50 percent less likely to drive a sale. The council's verdict is that the piece earns its claims: every figure traces to a named study, the unknown-delta story is told honestly as missing data rather than theft, and the quarterly-benchmark trend from 36 to 41 cents shows the fixes work. The takeaway for a media-buying decision-maker is concrete. The tax is a rate to be known and negotiated through log-level data and a shorter supply path; the waste is leakage to be cut with inclusion lists and curation. The advertisers moving the number are simply the ones who looked.
Comments