TL;DR: US advertising agencies are deploying AI across planning, bidding, and optimisation to reduce manual work, improve targeting, and respond to campaign signals faster than human teams can manage alone.

The State of AI in US Agency Buying

AI adoption in programmatic buying has moved from experimental to operational across many US agencies. The tools vary: some agencies use AI-assisted planning platforms, others have built internal tools, and a growing number are evaluating agentic systems where AI acts more autonomously on behalf of clients.

The common thread is speed. Programmatic advertising generates a volume of data that human analysts cannot process in the time available for decisions. AI systems can evaluate signals, adjust bids, and reallocate budgets in near real time.

Planning and Audience Modelling

Before a campaign launches, AI helps agencies model audiences more precisely. Traditional segmentation relies on predefined categories: age, location, intent signals. AI models can identify patterns across larger datasets and surface audience clusters that a human planner might not consider.

This does not replace the planning team. It changes their role. Planners work with AI-generated recommendations rather than building models from scratch. The planner's job becomes evaluating options and applying judgement about brand fit, context, and client objectives.

Bid Optimisation

Real-time bidding environments generate millions of auction opportunities per day. No human team can evaluate each one meaningfully. AI bidding systems assess the probability of a conversion, the value of an impression, and the current state of the campaign budget, then decide whether to bid and at what price.

Some agencies use black-box optimisation systems provided by DSPs. Others are building or buying tools that give their teams more visibility into how decisions are made. The demand for explainable AI in buying decisions is increasing, particularly from clients who want to understand where their money goes.

Campaign Monitoring and Reallocation

AI is also used to monitor delivery and flag anomalies. If a campaign is underdelivering against a particular placement type, or if frequency is building too quickly in a segment, an AI system can flag the issue and in some cases adjust automatically.

The level of autonomy varies. Some systems alert a human and wait for instruction. Others act within pre-set limits. The appropriate level of automation depends on the agency's risk appetite, the client's preferences, and the nature of the campaign.

The Move Towards Agentic Buying

The next stage beyond AI-assisted buying is agentic buying: AI agents that negotiate deals directly with sell-side systems, rather than simply bidding in open auctions.

In an agentic model, a buy-side agent might identify a publisher's available inventory, negotiate pricing and terms with a sell-side agent, and present a proposed deal for human approval. Once approved, the agent manages delivery within the agreed parameters.

This requires both sides to operate compatible systems and to follow shared protocols. The Agent Communication Protocol (AdCP), published by AgenticAdvertising.org, and the Agentic Advertising Marketplace Protocol (AAMP), published by IAB Tech Lab, are the two frameworks being developed to support this.

Marketplaces like Alkimi are building neutral infrastructure for these negotiations, giving buy-side and sell-side agents a shared DealSheet and an audit log of every agreement, without requiring either party to change their core systems.

What Agencies Should Be Doing Now

Agencies that are not yet evaluating agentic advertising should at least be monitoring it. The commercial advantage of faster, more accurate deal-making is significant, and early adopters will have a period of meaningful advantage before practices standardise.

Practically, agencies should:

- Audit which decisions in their current workflow are genuinely human-dependent and which are rule-following tasks that AI could manage. - Evaluate their DSP and SSP partners' readiness for agentic buying. - Begin internal education on how earned autonomy models work and what governance they require.

The agencies performing best with AI today are those that have been clear about where human judgement is essential and where automation adds speed without adding risk.