AI agents can negotiate digital out-of-home (DOOH) deals, and in some respects the channel suits them well: the inventory is finite and well described, with each screen having a fixed location, format and schedule, and programmatic DOOH already trades through deals. The difficulty lies in what a DOOH impression is. A screen does not know who looked at it. Delivery is recorded as a play in the media player's log, and the audience attached to that play comes from a model. An agent can only negotiate a DOOH deal safely if the deal record specifies the screens, the dayparts, the audience source and the play logs that count as proof.

TL;DR. DOOH inventory is sold by screen, location, daypart and share of a looping schedule, which gives an agent clear terms to negotiate. The counted delivery event is a play, while the impressions sold are an estimate built by multiplying plays by an audience figure from footfall, traffic or panel data. Two sellers can report the same plays as different impression totals if they use different audience sources, so the source and its version belong in the deal record. Proof-of-play logs are the main delivery evidence, and an agent can reconcile them against the deal only if the record states which screens, times and creatives should appear.

How is DOOH inventory different from other programmatic inventory?

DOOH inventory is a set of physical screens rather than a stream of individual user requests. Screens sit on roadsides, in stations, shopping centres, stores and gyms. Each has a fixed position, a format, a loop length (the cycle of ads that repeats through the day) and a number of slots within that loop.

When DOOH is traded programmatically, the screen's media player requests an ad for an upcoming slot, much as a web page requests an ad for a placement. The difference is that one play on a busy screen may be in front of many people at once, or no one. Targeting therefore works by place and time rather than by individual: venue type, location, daypart, and sometimes contextual triggers such as weather.

Media owners sell the same screens through direct bookings, programmatic guaranteed deals and private marketplaces, so an agent's first question on any offer is which kind of commitment is being made.

What do screens and dayparts mean in a DOOH deal?

Screens and dayparts are the core of the DOOH product. A screen list defines where the ad can run, ideally by the media owner's screen identifiers rather than a description such as "city centre roadside". Dayparts define when: morning commute, lunchtime, evening or weekends, in a named time zone.

The other terms that shape value are share of voice (the proportion of slots in the loop the advertiser holds), spot length and loop position. A buyer agent can trade these against each other within its mandate: a hypothetical brief might let it accept fewer screens in exchange for a higher share of voice during the morning commute, within a fixed budget. That trade is useful only if every variable ends up stated in the agreed terms.

Why are audience estimates not the same as impressions?

In web display, an impression is a counted ad serve to a device. In DOOH, the counted event is a play, and the impressions sold are a calculation: plays multiplied by an estimate of how many people were in a position to see the screen at that time. Media owners and their data partners build those estimates from sources such as footfall counts, traffic data, mobile location panels or camera-based counting, and the multiplier often varies by hour.

Suppose, hypothetically, a screen in a station carries an audience estimate of 12 people per play in the morning peak and 3 in the mid-afternoon. One thousand plays in each window would be reported as 12,000 and 3,000 impressions respectively. If the deal is priced per thousand impressions, the price depends as much on the multiplier table as on the plays. If the seller updates its audience model during the campaign, the same plays can be reported as a different number.

Impressions from different media owners are only comparable if the audience methodology is stated. Otherwise an agent may favour the most generous model rather than the best screens.

How do play logs work as delivery evidence?

A proof-of-play log is the media player's record of each ad it ran: screen identifier, creative identifier, start time and duration. It is the closest DOOH has to a delivery receipt, and it is the record a buyer, or a buyer's agent, should reconcile against the deal.

Play logs answer specific questions. Did the agreed creative run for its full spot length, on the agreed screens, within the agreed dayparts? Were any screens offline, and for how long? Machine-readable logs let an agent run that reconciliation routinely.

Play logs have limits. They come from the media owner's own systems, so they show what the player recorded, not what a passer-by saw, and independent checks such as site photography vary by market. A log is strong evidence of a play, and weaker evidence of an audience.

What must the DOOH deal record specify?

The deal record must contain every term that the play logs and audience reports will later be checked against. A working checklist:

Screens: the list by screen identifier, with venue type and location.
Dayparts: the windows, with time zone.
Share of voice and spot length: slots per loop and seconds per play.
Delivery unit: plays, or audience impressions; if impressions, the audience data source, methodology version and multiplier table.
Proof of play: the log fields, format and reporting cadence.
Outage tolerance: how much missed delivery is acceptable before a makegood applies.
Makegood rule: replacement plays on equivalent screens, within a stated period.
Creative approval: which party approves, by when; media owners commonly review creative against their own content rules and local advertising restrictions.
Price, spend cap and cancellation terms.

Creative approval and anything outside the mandate, such as a new venue type, should route to a person.

Where do the standards fit?

IAB Tech Lab's AAMP framework includes OpenProposal, which standardises how publishers define products and packages in RFP responses, so a media owner can describe a screen package in a form a buyer agent can compare with others. AdCP, the separate protocol published by AgenticAdvertising.org, names out-of-home among the channels it aims to coordinate. Neither removes the need for buyer and seller to hold the same account of what was agreed.

That shared account is the part Alkimi is built for. As a neutral agentic advertising marketplace working within IAB Tech Lab's AAMP framework, with its agent listed in the Agent Registry, it records the terms agents agree on a DealSheet that either side can retrieve independently, and people keep approval rights at the deal level.

Frequently asked questions

Can AI agents buy digital out-of-home advertising?

Yes. Programmatic DOOH already trades through deals, and an agent can negotiate screens, dayparts, share of voice and price within a mandate. Creative approval by the media owner usually remains a human step.

What is an impression in DOOH?

A DOOH impression is an estimate: a play multiplied by the number of people expected to be able to see the screen at that time, based on audience data such as footfall or traffic counts.

Why should the audience source be written into a DOOH deal?

Different audience sources and model versions produce different impression totals from the same plays. Fixing the source in the deal record keeps delivery reports comparable with what was bought.

What happens if a DOOH screen goes offline during a campaign?

That depends on the deal. A well-specified record sets an outage tolerance and a makegood rule, such as replacement plays on equivalent screens within a stated period.

Can AI Agents Negotiate Digital Out-of-Home (DOOH) Deals?