The programmatic report arrives on the first of the month, and it is beautiful.
Impressions in the tens of millions. A CPM lower than last quarter's. Viewability in the high sixties. A cost per click that seems, if anything, too good. Everyone nods, the invoice is paid, and a quiet question goes unasked: what did any of that actually buy?
If you have felt that question and not asked it, you are not being paranoid. The reporting layer of programmatic advertising was designed to be legible, not to be true, and the two are not the same.
Impressions are the least informative number on the page
An impression is a record that an ad was served. It says nothing about whether a human was present, whether the page had any reason to exist beyond hosting ads, or whether the placement was one the company would ever have chosen on purpose.
The Association of National Advertisers' 2023 study of programmatic supply chains found that a substantial share of impressions were landing on made-for-advertising sites: pages built to be filled with ads and visited by arbitraged traffic, with no audience anyone was trying to reach. Those impressions are cheap. They lower the average CPM. They also do nothing.
A low CPM, in other words, can be a symptom rather than a result.
What a real CPM is
Take the media cost. Remove the impressions that failed verification: invalid traffic, out-of-geo, non-viewable, made-for-advertising placements, and any inventory outside the company's own inclusion list. Divide the money by what is left.
That number is the real CPM, and it is often two to three times the reported one. It is the only CPM worth comparing across partners, months, or channels, because it is the price of impressions that could conceivably have mattered.
A desk that cannot produce this figure is reporting the price of a bag of impressions without opening the bag.
Rejection rate is the number to watch
The most useful single figure in programmatic is the rejection rate: the share of bid opportunities or served impressions the buy declined or discarded because they failed the company's standards.
A high rejection rate is not a problem. It is evidence that someone is choosing. A buy with a rejection rate near zero is accepting whatever the exchanges offer, and the exchanges offer what is cheapest to fill.
Ask for the rejection rate by partner and by reason. Then ask what the standards were. If nobody can answer the second question, there were no standards.
Six questions before the next invoice
What share of impressions passed verification, and what did the rest fail on?
Which domains and apps received the most spend, and would we have chosen them ourselves?
What is the real CPM, after removing everything that failed?
What is the rejection rate, and what rules produced it?
Which fees sit between our budget and the publisher, and what share of the dollar arrives?
Which outcomes, not clicks, can the buy be connected to, using tracking we control?
None of these is hostile. A competent desk will have the answers before you ask. A dashboard will not.
Why the fee structure decides the answers
Media bought on a percentage of spend rewards spending. Every incentive in that arrangement points toward more impressions at a lower average price, which is exactly the recipe for made-for-advertising inventory and a rejection rate of zero.
A flat fee per account changes the question the desk is answering. It stops being "how much can we place?" and becomes "did it work, and can we show it?" That is also why a desk should never front the media spend: the moment the desk is a lender, its interest in the volume of spend returns through the back door.
The numbers can be trusted. They just have to be the right numbers, produced by someone with no reason to flatter them.
lowob takeaway: Reported CPM, impressions, and viewability describe a bag of inventory. Real CPM, verification, and rejection rate describe what was inside it. Buy from a desk that reports the second set.
Selected sources: ANA: Programmatic Media Supply Chain Transparency Study (2023); MRC: Viewable Impression Measurement Guidelines