What is CPM?

A campaign spends A$3,200 and serves 400,000 impressions. CPM is (3,200 ÷ 400,000) × 1,000 = A$8.00. Serving those same impressions to a narrower, more contested audience might cost A$14.00 — the same thousand impressions, bought against different bidders.

Cost per mille is the price of a thousand ad impressions. Divide spend by impressions, multiply by a thousand, and you have the rate at which a campaign is buying attention-opportunities — not attention itself, and certainly not outcomes.

The unit is inherited. Print media sold advertising at a rate per thousand copies circulated, and digital advertising kept both the convention and the Latin abbreviation. The persistent confusion is the letter: mille is a thousand. Reading the M as million makes every CPM figure wrong by three orders of magnitude, which is a mistake worth naming plainly because it appears in real media plans.

An impression is a delivery event, defined by each platform's own rules about how much of the ad has to render and for how long. Those rules differ, so CPMs from two platforms are not measuring quite the same thing even before the audiences differ.

The spread across platforms is large and mostly structural. Gupta Media's tracker, which follows CPMs daily across its managed spend, put 2025 annual averages at US$4.82 for TikTok, US$4.99 for YouTube, US$8.19 for Meta and US$8.60 for Snapchat. Databox's median across more than 2,800 companies puts Meta at US$7.26, in the same neighbourhood from a different dataset.

LinkedIn sits in a different bracket entirely: Databox's B2B figure is US$26.91, roughly three times Meta's. That figure is from February 2023 and is the oldest number on this page — we cite it because it is the only LinkedIn CPM we could source to a credible publisher, and it should be read as an order of magnitude rather than a current rate.

The pattern behind the spread is supply and buyer value. TikTok and YouTube serve enormous impression volume, so supply is abundant and the clearing price is low. LinkedIn serves far fewer impressions to an audience whose job titles let advertisers attach large contract values to them, so the same thousand impressions clear far higher. Nothing about a low CPM says the impressions were good, and nothing about a high one says they were bad.

Competition. An ad auction prices against whoever else wants the same person. More advertisers chasing the same audience raises the clearing price with no change on your side at all.

Seasonality. Retail advertising concentrates into the last quarter, and CPMs across every consumer platform rise with it, peaking around Black Friday and Christmas before falling in January. A November CPM compared against a January CPM is comparing two different markets.

Audience size. Narrow targeting is expensive. A campaign restricted to a small, well-defined group is bidding against every other advertiser who defined the same group, and the auction has fewer people to spread demand across.

Placement. Feed, stories, reels, in-stream video and audience-network inventory clear at different prices. A campaign on automatic placements whose CPM moved may have changed nothing but the mix.

Expected performance. Most modern auctions rank on expected value, not bid alone — the platform's estimate of how likely the impression is to produce the outcome you asked for, multiplied by what you offered. A creative the model rates poorly effectively pays a surcharge, and it appears in the report as a higher CPM.

The most common error with this metric is treating it as something to minimise. The cheapest inventory in any auction is the inventory nobody else wanted, and there is usually a reason.

CPM sits at the top of a chain. It prices impressions; click-through rate turns impressions into clicks; conversion rate turns clicks into outcomes. The relationship is direct:

Cost per click = (CPM ÷ 1,000) ÷ CTR

An US$8 CPM at a 1% click-through rate produces an US$0.80 click. The same US$8 CPM at 2% produces a US$0.40 click. The impression price never moved; the creative did all the work. Conversely, a campaign that halved its CPM by drifting into low-attention placements and halved its click-through rate at the same time bought exactly nothing — and one of those two changes looks like a win in isolation.

This is why CPM is best read as a diagnostic beside its neighbours. On its own it tells you what you paid, never what you got.

CPM prices impressions without caring who saw them. A thousand impressions across a thousand people and a thousand impressions across two hundred people cost the same and are worth very different amounts.

Frequency — impressions divided by reach — is the missing variable. Nearly every campaign shows declining returns as frequency climbs: the second and third exposure still do work, the eighth mostly does not, and the audience's irritation is real even though no dashboard reports it. A campaign whose CPM held steady while frequency doubled got materially more expensive per useful impression, and the CPM column says nothing about it.

Small audiences make this worse quickly. A tightly targeted campaign with a large budget exhausts its audience and starts repeating, so it pays a high CPM to show the same people the same ad again.

Before treating a CPM movement as a problem, work out which direction the campaign moved in.

A CPM that rose because the campaign shifted toward a more contested, higher-intent audience is a price change and often a good trade — the click price and the cost per conversion may both have fallen. A CPM that rose with the audience and placements unchanged is competition or creative fatigue, and the fix is usually new creative rather than a new bid.

A CPM that fell deserves the same suspicion in reverse. Check whether reach grew, whether the placement mix shifted toward cheaper inventory, and whether the click-through rate fell by more than the CPM did. Cheaper impressions that convert less are not a saving.

The practical moves, in order: widen the audience so the auction has more people to price against; refresh creative before frequency climbs past the point where the model starts paying a surcharge; review the placement mix so you know what you are actually buying; and check reach alongside impressions so a rising frequency does not hide inside a flat CPM.

CPM is a price you agree to and effective CPM is a price you discover. On a buy sold per thousand impressions the two should almost match, and any gap is fees or a counting discrepancy. The difference matters on every other kind of buy: a campaign bought on clicks, views, installs or conversions never agreed an impression price at all, and the effective figure — spend divided by impressions delivered, times a thousand — is the only way to put it on the same axis as an impression buy. The practical division of labour is that CPM is a negotiation term and effective CPM is a reporting term. One is what a seller quoted; the other is what the media actually cost, and in a portfolio bought four different ways only the second one can be compared across it.

• Ask whether the CPM should come down at all. A CPM that rose because the campaign moved into a more contested, higher-intent audience is a price change, not a performance problem, and the click price may be lower than before.
• Widen the audience before you change the bid. Narrow targeting concentrates the auction on fewer people, and a small audience with several advertisers chasing it is expensive by construction.
• Check the placement mix. Feed, stories, reels and audience-network inventory clear at different prices, and an automatic-placement campaign whose CPM moved may simply have shifted mix.
• Watch frequency alongside CPM. Serving the same person a fifth impression costs the same as the first and is worth less, so a stable CPM with rising frequency is a rising real cost.