What is CPM (cost per mille)?

How CPM works

The calculation is simple: cost divided by impressions, multiplied by a thousand. A £5,000 campaign delivering two million impressions ran at a £2.50 CPM.

What the advertiser buys is exposure. Whether anyone clicked, converted or even looked is not part of the transaction, which is why CPM is the model that leaves all the performance risk with the buyer — and, correspondingly, all the upside.

eCPM is the number that matters

Effective CPM normalises any pricing model into the same unit, so a CPC buy and a CPM buy can be compared. It is total cost divided by impressions, times a thousand, regardless of how the deal was priced.

For publishers it is the core yield metric — the revenue a thousand impressions actually produced across every demand source. For advertisers it is how you tell whether a CPA deal that felt expensive was, in exposure terms, competitive.

On AI surfaces the equivalent normalisation is often per query rather than per impression, because a query is what the publisher produces and only some fraction carry advertising — see revenue per query.

Why AI CPMs are high

• Supply is deliberately capped. Publishers limit advertising to protect the product, so inventory does not expand with usage.
• One placement per conversation, not several per screen.
• Intent is stated, not inferred, which makes each exposure worth more — see intent density.
• No arbitrage. The moment cannot be reached more cheaply anywhere else.

The trap

A low CPM is not a good deal, and this is the most common error in media buying. Cheap impressions are cheap because they are worth less: unviewable, fraudulent, or served to people with no relevant intent.

The comparison that matters is cost per outcome. High-CPM AI inventory frequently produces a lower cost per outcome than cheap display, and a buyer optimising CPM downwards will systematically move budget toward inventory that does nothing. Judge it on ROAS or on measured lift, never on the CPM alone.