Everyone in advertising knows ad fraud exists. It comes up on conference panels, in trade headlines and in industry reports. Yet when I sit down with marketing teams, the conversation tends to land in the same place. It's a real problem, just not ours. Our agency handles it. Our verification vendor would have flagged it. And besides, look at our CPMs.
That last line worries me most.
A comfortable number
CPM is a comfortable number. It's simple, easy to compare, and it falls nicely when buyers negotiate hard or move budget into open programmatic exchanges. A plan that delivers twice the impressions for the same money looks like good stewardship.
But a CPM only tells you what you paid for a thousand ad impressions. It tells you nothing about who, or what, was on the other end.
It’s like a market stall selling oranges at half the price of the greengrocer across the road. A bargain, until you open the crate and find that only one orange in ten is fit to eat. You didn't buy cheap oranges. You bought very expensive ones, packed with rot.
A low CPM can be that crate. On its own, it's a vanity metric.
Price the humans, not the impressions
Dr Augustine Fou, the independent ad fraud researcher behind FouAnalytics, has spent more than a decade measuring who actually sees digital ads. His answer to the CPM problem is a metric he calls hCPM, or human CPM (FouAnalytics, February 2024).
The arithmetic is simple. Take what you paid per thousand impressions and divide it by the share of those impressions that reached humans. Fou explains: "Say you are running programmatic ads and paying a nominal CPM of $3. If only 10% of those ads are shown to humans, you need to buy 10X more ads to achieve the same effect as if all of your ads were shown to humans." Divide $3 by 0.10, and you're effectively paying $30 per thousand humans.
Is 1% far-fetched? Fou's own sample data suggests not. He compared 22 programmatic campaigns with 11 buys from real, well-known publishers, each in the range of roughly 35,000 to 49,000 impressions. In none of the programmatic campaigns did the share of impressions shown to humans reach 1%. Among the publishers, it ranged from 28.5% to 80.5%. Fou’s summary is shown below:
Put two lines side by side, and the gap is stark. One programmatic campaign delivered 47,260 impressions and reached 194 humans. One publisher buy delivered 45,000 impressions and reached 28,667. "There's literally two orders of magnitude difference," Fou writes.
His challenge to buyers is blunt: "Would you prefer to buy a ton of low-priced ads, with only 0.5% of impressions shown to humans, or would you prefer to buy a far smaller quantity of ads, with most of the ads shown to humans? Yeah, you get what you pay for."
He goes further still: "Even if you have 100X higher CPMs, you can buy far fewer ads. You will still save costs."
Why it never seems to happen to us
If the gap is this wide, why do so many smart teams believe they're immune?
Partly because the reports look fine. Impressions are delivered, CPMs are on target, and the dashboards are green. Bots don't write complaint letters. Unviewable ads don't appear as blank spaces in a campaign report. The waste is invisible precisely because every number we're used to checking looks healthy.
Partly, too, because the incentives point the wrong way. A team that halves its CPM gets praised. A team that doubles its CPM to reach real people has some explaining to do, even if it sells more. As Fou notes, "Advanced ad buyers today optimise for 'humanness' and human CPM, as opposed to nominal CPM." By implication, many others are still chasing the cheapest number on the sheet.
The evidence that quality pays
This isn't just one researcher's view. The Association of National Advertisers (ANA) Q4 2025 Programmatic Transparency Benchmark measured how much programmatic spend turns into impressions that are fraud-free, measurable, viewable and not on made-for-advertising sites. Advertisers with disciplined quality governance converted 56.7% of their spend into impressions that met that bar. Lower-performing advertisers managed 37.5%, which means most of their money failed the test.
The trend isn't reassuring either. The share of advertisers clearing the 50% mark fell from 50.0% in Q3 to 43.8% in Q4.
The most telling finding is about outcomes. Advertiser case studies in the benchmark reported cutting cost per conversion by nearly 40% after optimising towards quality-adjusted metrics rather than CPM alone, even where nominal CPMs rose. Paying more per impression, and less per customer. That's the human CPM argument playing out in real campaigns.
Three questions for your next media review
None of this requires a revolution. It requires a different first question. Before you next sign off a plan, ask:
1. What share of our impressions were actually shown to humans?
2. What is our cost per thousand humans, site by site and app by app?
3. When our CPM went down, what happened to our cost per conversion?
If nobody can answer the first two, the third probably isn't telling you what you think it is.
The cheapest impression in your plan may still be a good one. But until you know how many people stood behind it, you don't know what it cost.
If you'd like to see what your own cost per human looks like, InMail me about a free audit.
.gif)
