What is Impression?

A campaign spends A$4,000 at a A$6.50 CPM. Impressions are (4,000 ÷ 6.50) × 1,000 = 615,384. If those impressions reached 120,000 people, the average frequency is 615,384 ÷ 120,000 = 5.1 — the same delivery described as volume, as people, and as repetition. All three come from the same count and only the first one appears on the invoice.

An impression is one delivery of an ad. The platform served it, a screen was asked to render it, and the count went up by one.

That is the whole definition, and its narrowness is the point of the rest of this page. The count makes no claim that a person looked at the ad, that the ad finished loading, that it appeared in the visible part of the page, or that the person was awake. It records that an opportunity to be seen was created and paid for.

Because it is the most basic unit of delivery, it ends up underneath almost everything else. Click-through rate divides clicks by impressions. Cost per thousand impressions prices them directly. Effective cost per thousand converts any buy into their units. Frequency divides them by people. View rate divides views by them. A misunderstanding here propagates a long way.

Platforms differ in the exact moment the count fires. Some record an impression when the ad is served by the server, some when the creative begins rendering in the browser or app, some when a defined share of it has rendered. Video is treated differently again, particularly autoplay, where the ad may begin playing without any deliberate action.

These differences are small enough to ignore inside one platform and large enough to make cross-platform impression comparisons approximate. When two systems report on the same campaign — an ad server and a platform, for instance — a discrepancy of a few percent is routine. A large or growing gap is worth investigating, because it moves every derived metric that divides by impressions.

The gap between an impression and a person seeing an ad is the reason viewability exists as a separate discipline.

An impression can be counted for an ad below the fold that the person never scrolled to. For one in a background tab. For one at the extreme edge of a viewport. For one that appeared for a fraction of a second during a fast scroll. For one on a page that loaded and was abandoned before anything rendered fully.

None of this is fraud, and it is not a defect in the count. Delivery and attention are different things and the industry measures them separately for good reason. An advertiser comparing two sources at the same cost per thousand impressions, without knowing their viewable shares, is comparing prices for different goods.

The most useful single companion to this count is reach, because the two together produce frequency.

Six hundred thousand impressions delivered to a hundred and twenty thousand people is a campaign showing each person five ads. The same six hundred thousand delivered to half a million people is a campaign showing most of them one. Both cost the same. They are completely different campaigns, and the difference is usually a consequence of audience size and delivery settings rather than a deliberate decision.

Any report that leads with impressions and omits reach is concealing which of those two campaigns is running. For direct-response work the concealment often does not matter. For anything where being known by more people is the objective, it is the whole question.

Impressions at a fixed budget move inversely with the price of inventory, which means the count is a fairly direct read on where delivery is going.

A rise in impressions at flat spend means the average price fell. That happens when the audience broadens, when placements shift toward cheaper inventory, when competition in the auction eases, or when a bid strategy changes. None of those are inherently good or bad, and none of them are visible in the outcome metrics straight away.

The habit worth building is to treat a sudden impression increase as a question rather than a result. Automated bid strategies in particular will happily buy a great deal of cheap inventory in pursuit of a conversion target, and the impression count is where that shows up first.

Pricing. It is the unit the auction charges in, directly or derived, and therefore the unit in which media cost is compared.

Denominating rates. Every rate metric in advertising needs it, and the quality of those rates depends on the count being consistent.

Sizing a plan. Working backwards from a required number of conversions through conversion rate, click-through rate and price gives the impressions a plan needs, which is how a budget gets sized before anything is bought.

Detecting delivery drift. As above: the count moves before the outcomes do.

It is not a measure of success and it is not a measure of exposure. A campaign that doubled its impressions has bought more of something; whether that something was worth having is a question the count is structurally incapable of answering.

Presented on its own — in a report, a case study, a board slide — a large impression number is close to content-free. It becomes informative the moment it is put beside how many people it reached, how much of it was viewable, what it cost per thousand, and what happened afterwards.

The one place the raw number carries a decision is at the front of a media plan, before anything is bought.

Start from the outcome the plan needs: a number of sales, leads or installs. Divide by the rate at which clicks produce that outcome to get the clicks required. Divide by the click-through rate the creative and placement realistically achieve to get the impressions required. Multiply by the price per thousand to get the budget.

Each of those three rates is an assumption, and stating them explicitly is the value of the exercise. A plan that needs eleven million impressions on an audience that can only generate four million in the flight window has a problem that no amount of optimisation will solve, and the arithmetic finds it before the money is committed rather than in the third week.

Run in reverse, the same chain sizes what a fixed budget can be expected to produce, which is a more useful answer than a target somebody picked.

Impressions count deliveries; reach counts people. A campaign with 600,000 impressions against 120,000 people has served each person five ads on average, and a campaign with 600,000 impressions against 500,000 people has served most of them one. The two campaigns cost the same and are doing completely different things. Impressions are what the auction charges for and what every rate metric divides by, which is why they dominate reporting. Reach is what a launch, a brand campaign or a market-entry plan is actually buying, and it is the harder of the two to grow because it requires finding new people rather than showing the same ones another ad. Read them together: the ratio between them is frequency, and frequency is where most delivery problems become visible.

• Treat the count as a denominator rather than an achievement. More impressions at the same spend means cheaper inventory, which is not automatically better inventory.
• Read it beside reach in every report. The same impression count spread over ten times the people is a different campaign, and the delivery settings that produced it were different too.
• Check the viewable share before comparing two sources. An impression counted below the fold and one that held the screen cost the same and are not the same event.
• Split by placement before drawing conclusions. Feed, stories, in-stream and audience network deliver impressions at prices that differ by multiples.
• Watch the count when a bid strategy changes. A sudden increase in impressions at flat spend usually means delivery moved to cheaper inventory rather than that the campaign improved.