A small business can use an AI agent to buy TV by deciding the goal first, giving the agent a written mandate with hard limits, keeping a person's approval on every deal, testing the agent before real money moves, and checking a shared record of the agreed terms afterwards. No trading desk is needed. Software agents and open standards now do much of a desk's legwork, but the judgement stays with the business.

TL;DR. Small and medium advertisers can now reach TV and connected TV (CTV) through AI tools built into buying platforms, through buyer agents that negotiate with seller agents over open protocols, or through agent marketplaces. The eight steps below run from setting the goal to scaling up, with a person approving each deal until the agent has earned more room. The two steps most often skipped are testing the agent before it spends anything and checking the deal record once it has.

What options do small businesses have for AI-agent-driven TV buying?

Demand from smaller advertisers is already there. Digiday reported in February 2025 that one agency, Collective Measures, estimated ten of its clients had used CTV to run TV advertising for the first time. There are three broad routes for adding an agent, and they differ in how much the agent decides.

The first route is AI built into existing buying tools, for example to make a TV-ready ad or set up a campaign. When Magnite announced its acquisition of streamr.ai in September 2025, it described the technology as AI tools that make CTV advertising accessible to small and medium-sized businesses.

The second route is a buyer agent negotiating directly with a seller agent over an open protocol. Since version 2.0, IAB Tech Lab's Agentic Advertising Management Protocols (AAMP) have supported programmatic guaranteed deals, preferred deals, private marketplaces, the open marketplace and linear GRPs. AgenticAdvertising.org, which publishes the Ad Context Protocol (AdCP), records the first agent-to-agent media buy on 16 October 2025, using inventory from LG Ads with human oversight.

The third route is a marketplace where buyer and seller agents meet under shared rules. Alkimi is one worked example: a neutral marketplace that works within IAB Tech Lab's AAMP framework today and is being built to be AdCP compatible, where each agreed deal is written to a DealSheet held by both parties and a person keeps approval rights at deal level.

Named providers in one market are covered in how UK small businesses can advertise on TV without a trading desk, and the general trade-offs in do you need a trading desk to buy CTV?

Step 1: What goal and audience should the agent buy for?

Start with one outcome the business can measure, such as visits to a shop, calls to a branch or sign-ups on a website. AgenticAdvertising.org puts the risk plainly: "Without governance, agents optimize for whatever metric they're pointed at." So write the goal as a sentence a stranger could check later.

Then define the audience in terms a seller can deliver: regions or postal areas, household characteristics, and the hours or programmes that fit. Name the places the business does not want to appear, because exclusions are easier to enforce when they are written down before negotiation than argued about afterwards.

Step 2: What goes in the agent's mandate?

A mandate is the written instruction set an AI buying agent works to: budget, flight dates, audience, frequency cap, acceptable sellers and the minimum terms any deal must include. A fuller template is in what a CTV media buying mandate should contain.

The minimum terms matter more than the budget line. For TV and CTV they usually cover what counts as a view, the viewability floor, the completion rate, the make-good if delivery falls short, and how and when reporting arrives. Without them, an agent can complete a buy that leaves the business barely protected.

Step 3: Where should the approval threshold sit?

The approval threshold is the point above which the agent must stop and ask a person. For a first campaign it should sit at zero: the agent can search, compare and draft, but every deal goes to a named person before it binds. Earned autonomy describes five stages (Observe, Recommend, Draft, Human-approved action and Bounded automatic action), and a new agent should start no higher than the fourth, as set out in the five stages of earned autonomy in AI media buying.

Gartner figures reported by Adweek put highly autonomous deployments at less than 5% of all marketing use cases in 2025, and Gartner predicts that only 15% of agentic AI deployments will be highly autonomous by 2028.

Each side of a deal can set its own limit. AgenticAdvertising.org gives the example of a seller that pre-approves any order within 10% of a buyer's usual weekly spend and reviews everything else by hand, while the buyer approves every order personally.

Step 4: How should the business benchmark or test the agent?

Test the agent on realistic briefs before it touches a live budget, and judge it on what it secured, not only on whether it finished. In the Concourse Bench v1 study, run in a controlled market, 55 of 96 registered attempts by AI buyer agents ended in a completed buy. Across the 110 contracts signed in those completed buys, 95 promised a viewability floor below 70%.

"The agent completed the buy" is the start of the review, not the end. Ask any agent provider for test results that show the terms secured, such as viewability floors and make-goods, alongside the completion rate.

Step 5: Why practise against seller agents before going live?

A buyer agent behaves differently against a seller agent that holds firm than it does in a demo. Practice rounds show whether the agent keeps to its mandate under pressure, walks away when it should, and escalates to a person at the threshold set in step 3.

IAB Tech Lab publishes open-source Buyer and Seller Agent SDKs as reference implementations, so a buyer agent can be run against a reference seller agent. Before any live negotiation, the business should also confirm who it is dealing with: the IAB Tech Lab Agent Registry, part of AAMP, is free to join and provides a framework for agent identity, verification and disclosure.

Step 6: How should the first live deal be run?

Keep the first deal small, short and with one seller, and have a named person approve it before it binds. The aim is to test the whole chain, from brief to signed terms to delivery report, rather than to hit a sales target.

Standards are taking manual work out of this stage. IAB Tech Lab's AAMP 3.0 introduced OpenProposal, a specification that lets sellers respond to buyer briefs in an agent-readable format, and sets out a lifecycle of publish, discover, compare, qualify, negotiate, bind and execute, and report. The OpenProposal public comment period ran from 22 September to 22 October 2026.

Step 7: What should the deal record show?

The deal record should show the terms both sides agreed, at the moment they agreed them: audience, flight, frequency cap, viewability floor, completion rate, make-goods and reporting. Each side should be able to retrieve it independently, so neither has to rely on the other's copy when delivery is checked.

Compare the record with the mandate from step 2 before the campaign starts. A missing or weakened term means the approval in step 6 was not strict enough.

Step 8: How should the business review results and decide whether to scale?

Review delivery against the deal record, not against the agent's summary of the deal. Check whether the agreed audience, frequency and viewability were met, whether any make-good was triggered, and whether the outcome from step 1 moved.

Scale one variable at a time: a bigger budget, a second seller or a higher approval threshold, but not all three together. Move the agent up a stage of autonomy only after several deals in which its drafts needed no material correction.

What does this look like for a hypothetical small business?

This example is hypothetical and every figure in it is invented. A regional chain of four furniture shops wants more weekend visits during a four-week spring sale. Its mandate names three regions, households likely to be moving home, three exposures per household per week, an 80% viewability floor and a make-good for any shortfall, with the operations manager approving every deal.

In practice rounds, the agent accepts a 60% viewability floor from a firm seller agent in two of five attempts, so the business tightens the instruction to walk away below the floor. In the live run, the agent returns two proposals from streaming publishers, the manager approves one, and the deal record shows the floor, cap and make-good as written.

After the flight, the business compares delivery with the record and weekend visits with the previous spring, and only then decides whether to add a second publisher.

Frequently asked questions

Can a small business buy TV advertising without a trading desk?

Yes. Broadcasters, streaming services and self-serve platforms sell to smaller advertisers directly, and AI agents can now do much of the searching and negotiating a desk once did.

Does an AI agent replace the person who buys the media?

No. The person writes the mandate, sets the approval threshold and signs off each deal, while the agent does the searching, comparing and drafting.

What is a media buying mandate?

A mandate is the written set of rules an AI buying agent must follow, covering budget, audience, dates, sellers and minimum deal terms. It is the main control a business has over what the agent can agree to.

Which standards do agent-bought TV deals use?

IAB Tech Lab's AAMP covers agent workflows from proposal through to execution and reporting, building on existing standards such as AdCOM, OpenDirect and the Deals API. AdCP, published by AgenticAdvertising.org, is an open protocol that lets agents discover, plan, buy, sell and measure media.

Where can a small business practise these steps?

Alkimi, an enterprise platform for agentic advertising, covers steps 4 to 7 with three products: Concourse, the live benchmark behind the study in step 4; the Gym, where buyers and sellers build their own agents, set the rules they work to and spar against counterparty agents before any real money moves; and the Marketplace, where each deal is written to a timestamped DealSheet either party can retrieve independently. The Gym is available now, and enquiries go through the Concourse contact form.

The company also filed comments and change requests during the OpenProposal 3.0 public comment period. Whichever route a business chooses, the eight steps hold: a clear goal, a written mandate, a person approving and a record both sides can check.

How a Small Business Can Use an AI Agent to Buy TV: A Step-by-Step Guide