Attention per mille, usually shortened to APM, is the total number of seconds people spend looking at an advertisement for every 1,000 impressions served. It takes the logic of CPM (cost per mille, the price of 1,000 impressions) and swaps the unit being counted: instead of ads delivered, it counts attention received. A score of 2,000 APM means that 1,000 impressions generated 2,000 seconds of eyes-on viewing between them, or two seconds per impression on average.

The metric exists because an impression, and even a viewable impression, says nothing about whether anyone looked. Under the Media Rating Council (MRC) standard, a display ad is viewable when half its pixels are on screen for one continuous second, two seconds for video. That threshold confirms an opportunity to see. APM tries to estimate what happened inside that opportunity, and to express the answer in a unit that can be set against price.

How the calculation works

The most widely cited formula comes from Lumen Research, the London-based attention company whose models sit behind many APM figures in the market. According to a Lumen glossary published by the demand-side platform Eskimi, APM equals the share of impressions viewed multiplied by the average view time, multiplied by 1,000. An equivalent version divides total view time by measured impressions and scales the result to 1,000.

A worked example shows the two inputs at work. Take 1,000 impressions of a mobile video unit. If 40% of them are actually looked at, and those that are looked at hold the eye for an average of 2.5 seconds, the campaign has earned 1,000 attentive seconds. Its APM is 1,000. Raise the view rate to 60% and the average to five seconds, and APM climbs to 3,000.

That split matters. View rate captures breadth: how many impressions drew any gaze at all. Dwell time captures depth: how long the gaze stayed. APM multiplies the two, so a format seen briefly by many people and a format seen at length by few can land on the same score.

Because APM is a volume measure rather than a rate, it adds up across a campaign. Teads' help documentation describes it as looking "at all impressions," in contrast with eyes-on dwell time, which describes only impressions that were viewed.

Price enters through a companion figure. Lumen defines aCPM as total cost divided by total view time in seconds, multiplied by 1,000, which yields the cost of 1,000 attentive seconds. A banner bought at a EUR 2 CPM with an APM of 250 costs EUR 8 per thousand attentive seconds. A 15-second television spot bought at a hypothetical EUR 30 CPM with an APM of 5,000 costs EUR 6. On that arithmetic the expensive unit is the cheaper source of attention, which is the comparison APM was designed to make possible.

Where the numbers come from

Almost no APM figure is observed directly. Vendors recruit consenting panellists, record their gaze through webcams or dedicated eye-tracking hardware, and train machine learning models that predict attention from signals available at scale: time in view, scroll speed, ad size, screen position and page clutter. According to Teads, Lumen's panels include 1,000 users in the United States and 1,000 in the United Kingdom. The model then scores every impression in a campaign without a live panel.

On the buy side, APM appears in post-campaign reports, in guaranteed deals and increasingly as a delivery commitment. Lumen measurement became billable inside The Trade Desk through a native billing integration announced on March 25, 2024, switched on per ad group. On the sell side, publishers and ad networks use APM to argue that high-impact or interactive formats deserve premium pricing. InMobi's FRAME Attention product, launched on July 24, 2026, guarantees an 8,000 APM floor across in-app video and connected television (CTV), equivalent to eight attentive seconds per impression across full delivery.

Origin and evolution

The idea of attentive seconds as a cross-media currency predates the acronym. On October 18, 2019, Lumen published a paper by Andrew Nappin titled "Attentive seconds: one currency to rule them all?" It combined three rates, viewability, likelihood to see and view time, into a single figure. On a desktop MPU (a 300 by 250 pixel display unit), 57% of 1,000 impressions were viewable, 8% were looked at and each look lasted 1.3 seconds, producing about 110 attentive seconds. A quarter-page press advertisement produced around 1,400 and television around 6,000. Nappin noted that television delivered about four times the attention of press, so paying ten times the CPM would not be a good deal.

Agencies then built pricing around the same idea. In January 2022 Doug Rozen, Americas chief executive of Dentsu Media, described an "effective attention cost per thousand", or eaCPM, to Beet.TV as a replacement for the traditional CPM in valuing video formats. Vendors formalised the acronym in their documentation. Adnami's help centre, in a page dated February 8, 2024, defined APM per 1,000 viewable impressions and introduced eAPM, effective attention per mille, per 1,000 purchased impressions.

The industry's rulebook arrived later. The MRC and IAB (Interactive Advertising Bureau) released draft attention measurement guidelines on May 12, 2025, and finalised them on November 12, 2025. The final document does not define APM or any per-thousand attention metric. It governs the inputs: attention must sit on top of viewable impressions, invalid traffic must be filtered, models need a robust deterministic dataset, and attentive time may be accumulated on a non-continuous basis within a session. Attention measured without viewability may be reported only as non-standard and diagnostic.

Why it matters for marketers

APM gives brand advertisers a way to compare a six-second mobile banner and a 30-second CTV spot on one axis. YOC and Lumen reported on September 2, 2025 that YOC's interactive Mystery Ad scored 5,634 APM against 5,178 for a 15-second television spot, while standard mobile banners scored 244 and skippable YouTube video about 2,344.

Teads uses the same unit. Its CTV HomeScreen campaigns average about 5,300 APM, according to Teads, after Lumen measurement went live on that inventory on May 6, 2026. Lumen predictions, expressed as active attention seconds per thousand impressions, reached Netflix inventory in five European markets in March 2026.

APM is one of several attention signals now plugged into buying systems. Adelaide's AU score became a Custom Bidding input in Display & Video 360 from mid-July 2026, and Viant closed its acquisition of TVision on May 5, 2026 to price CTV on an attention-adjusted CPM. The broader category is mapped in Explaining attention metrics.

Limitations and disputes

No standard definition. YOC's own documentation, updated in April 2026, says APM is "not a standardized KPI." Lumen's APM counts predicted seconds per measured impression. Adnami's APM counts proprietary "attention points" per 1,000 viewable impressions, which makes its figures structurally different. In Adnami's example, a campaign where 70% of impressions were seen shows an eAPM 30% below its APM. Two vendors can report the same campaign at very different numbers.

Modelled, not observed. Most scores are predictions extrapolated from panels of a few thousand people. The MRC guidelines require models to be validated against fresh deterministic data, but accreditation for specific APM products had not been announced as of October 2026.

Seller-commissioned evidence. InMobi's claims that its CTV inventory achieved 9,593 APM against a 7,286 benchmark came from testing the company commissioned, and the benchmark had no stated source. Teads' 5,300 figure has not been independently validated.

Duration is not effect. An average hides the distribution. A score of 2,000 can mean many glances or a few long stares, and the two do different work. A Kantar analysis of 873 campaigns, combining its LIFT+ database with Lumen data, found that channel attention levels did not directly track brand contribution or cost-effectiveness. Peter Buckley, connection planning director at Meta, wrote that "some advertisers are buying too much attention." The MRC guidelines say attention "should not be considered or used as a measure of outcomes." Long formats also enjoy a built-in advantage, since a 30-second spot has more seconds to fill than a static banner.

Not the same as

CPM is a price per 1,000 impressions served. APM is a quantity, attention per 1,000 impressions, and carries no price.

vCPM (viewable CPM) prices 1,000 viewable impressions. Google opened viewability-based CPM buying across its display network on December 12, 2013, using the 50% of pixels for one second threshold. Viewability is a precondition for APM, not a measure of it.

aCPM or eaCPM is the cost of 1,000 attentive seconds, derived by dividing cost by APM. Adnami calculates it as CPM divided by eAPM, multiplied by 1,000.

Attention scores such as Adelaide's AU, a 0 to 100 index, or IAS Quality Attention rank the likelihood of attention rather than counting seconds. Cumulative attention, which adds up exposure across a campaign and over time, is a related but separate concept.

Recent developments

The pace picked up through 2026. Adelaide's attention segments entered Amazon DSP as pre-bid targeting on June 10, 2026. InMobi's 8,000 APM floor in July was the clearest sign yet of APM moving from a reporting metric to a delivery guarantee, although the company did not say what remedy applies if a campaign falls short. TVision reported on August 25, 2026 that only nine of more than 200 daily video viewing minutes were attentive ad time, a panel figure that frames how scarce the counted seconds are. Whether APM settles into a shared currency depends on a common definition of the second being counted.

Timeline

  • October 18, 2019 - Lumen Research publishes Andrew Nappin's paper proposing attentive seconds as a cross-media currency.
  • January 10, 2022 - Dentsu Media's Doug Rozen describes eaCPM, an effective attention cost per thousand, as an alternative to CPM.
  • February 8, 2024 - Adnami documents the gap between APM and eAPM.
  • March 25, 2024 - Lumen measurement becomes billable natively inside The Trade Desk.
  • July 15, 2024 - Kantar releases its 873-campaign study linking attention to brand outcomes.
  • May 12, 2025 - MRC and IAB release draft attention measurement guidelines for public comment.
  • September 2, 2025 - YOC and Lumen publish APM comparisons between high-impact mobile formats and 15-second TV spots.
  • November 12, 2025 - MRC and IAB finalise attention measurement guidelines, without defining APM.
  • March 5, 2026 - Lumen announces attention measurement on Netflix ads in five European markets.
  • May 6, 2026 - Teads brings Lumen measurement to CTV HomeScreen globally, reporting about 5,300 APM.
  • July 24, 2026 - InMobi launches FRAME Attention with an 8,000 APM floor.

Summary

Who: Attention vendors such as Lumen Research and Adnami calculate APM; sellers including Teads, YOC and InMobi publish and guarantee it; agencies and advertisers use it to compare formats and set prices.

What: A count of the seconds of human attention an ad earns per 1,000 impressions, usually the share of impressions viewed multiplied by average view time and scaled to 1,000, paired with aCPM, the cost of 1,000 attentive seconds.

When: Lumen proposed attentive seconds as a cross-media currency in October 2019; vendors documented APM by 2024; the MRC and IAB attention guidelines of November 12, 2025 govern its inputs but not the metric itself.

Where: In post-campaign reports, DSP fee integrations, publisher format pitches and, since July 2026, delivery guarantees across display, social, in-app video and CTV.

Why: Impressions and viewability confirm an ad could be seen, not that anyone looked. APM puts a number on looking so that cheap and expensive formats can be compared, though vendor definitions differ, scores are modelled and attention does not equal effect.