The cheap impression illusion.

CPM is the metric the programmatic industry built its entire sales pitch around. Low cost per thousand impressions. Massive scale. Thousands of sites, millions of users, fractions of a euro per ad served.

It is also one of the most misleading numbers in marketing.

Not because the arithmetic is wrong — CPM is calculated correctly. But because CPM measures the cost of serving an ad, not the cost of it being seen by a real person. And between those two things, a significant portion of most programmatic budgets quietly disappears.

When spend moves through an open programmatic exchange, it passes through a chain of intermediaries before reaching a publisher. Each takes a cut: the demand-side platform, supply-side platforms, data providers, verification vendors. By the time money reaches the entity that actually has an audience, a substantial portion is already gone.

How much? The ANA tracked $123 million in real ad spend and found that in 2023, only 36 cents of every dollar entering a DSP ended up as a viewable impression on quality inventory. By 2024, after significant industry pressure, that improved to 43.9 cents. Progress — but still less than half.

Adalytics research found that in some individual programmatic transactions, the combined platform take reaches 35% of the advertiser’s bid. In extreme cases, the publisher — the one with the actual audience — received just 2 cents of every euro spent. The advertiser had no visibility into which transactions were which.

This is not a flaw in an otherwise transparent system. It is the system working as designed. Opacity benefits the intermediaries. Clarity would not.

Beyond the fee problem, there is a more fundamental quality issue: a large share of the inventory available in the open exchange serves no real audience.

Made-for-advertising (MFA) sites are built not to inform or entertain anyone, but to generate programmatic revenue. Sensationalist headlines, recycled content, bot-inflated traffic. Their business model is capturing CPM budgets from advertisers not paying attention to where their ads land.

At the peak of the problem, in mid-2023, MFA sites accounted for nearly 30% of all open web programmatic auctions. One in five impressions — going to sites with no genuine audience.

Industry action has brought that number down sharply. Median MFA spend fell below 1% by Q1 2025. But the ANA’s same benchmark found that 25% of marketers still had MFA exposure between 5% and 28.7% of their programmatic spend.

MFA sites score higher on viewability than legitimate inventory and are more measurable. On paper they look like quality placements. But their invalid traffic rates are nearly six times higher than clean inventory. The surface metrics that make them look good are exactly the ones that can be gamed.

One of the most striking findings from the ANA’s research was not about fraud or fees. It was about unnecessary scale.

The average programmatic campaign tracked in the study ran across 44,000 individual websites. The ANA’s analysis found that roughly 500 well-selected sites would have been sufficient to reach approximately 95% of the same audience. The other 43,500 sites were not adding reach — they were adding brand safety risk, supply chain complexity, and opportunities for waste, while contributing nothing of measurable value.

Those 43,500 extra sites exist because the open exchange optimises for the metric it is given. When that metric is CPM, it finds the cheapest impressions available. The cheapest impressions available are, almost by definition, on the lowest-quality inventory. The system is not broken. The brief is.

The right way to compare programmatic inventory is not nominal CPM — the price per thousand impressions served. It is effective CPM: the cost per thousand impressions genuinely delivered to a real person, in a real attention environment, free from fraud.

Once you adjust for viewability rates, invalid traffic, supply chain fees, and brand safety exposure, the economics look very different. A €2.50 display CPM on the open exchange, after accounting for everything that is lost in the chain, has an effective CPM closer to €5.40. A €44 premium CTV placement that delivers near-100% viewability and 95%+ completion has an effective CPM of approximately €44.70.

On that basis, premium CTV is not seventeen times more expensive than open-exchange display. It is roughly eight times more expensive on an effective basis — and it delivers near-100% viewability, a lean-back audience with sound on, and near-zero brand safety risk.

PMP (private marketplace) CPMs average around 2× open-exchange levels. On an effective CPM basis — adjusted for viewability and fraud — they are frequently price-competitive with open-exchange inventory, and come with known publisher environments, significantly lower fraud exposure, and guaranteed brand-safe adjacency. The headline CPM premium is not a cost. It is the price of knowing where your ad ran.

There is one more dimension to the cheap impression calculation that most media plans ignore entirely: the brand safety exposure that comes with buying at scale across unknown inventory.

When your ad runs on 44,000 websites, you do not know what those websites contain. The open exchange does not guarantee adjacency. A family brand next to extremist content, a financial services ad on a misinformation site — these are documented occurrences from campaigns that optimised for low CPM without specifying where that CPM should land.

“Everyone says they are supporters of transparency, until they’re the ones asked to be transparent.”

— ANA PROGRAMMATIC SUPPLY CHAIN TRANSPARENCY STUDY, 2023

The cost of a brand safety incident is not a CPM. It is the erosion of consumer trust that took years and significant investment to build. No programmatic fee structure prices that in. But it is real, and one bad placement can undo far more brand equity than the CPM saving it generated.

Good programmatic buying starts with a supply path question, not an audience question. Who are the publishers? How many intermediaries sit between your bid and their inventory? Can you name them?

It then applies a curation lens: a defined allowlist of publisher environments where you are willing to run, purchased through direct deals or tightly managed PMPs. It does not mean abandoning scale — it means buying the scale that actually matters.

Finally, it measures on effective CPM, not nominal CPM. If your reporting does not distinguish between served impressions and viewable impressions on quality inventory, you are flying blind — and the open exchange is making that choice for you.

The cheap impression is not cheap. It is the most expensive impression you can buy, because it costs money while doing nothing.