How agentic AI’s machine muscle and the thinking power of media indies can level the playing field
Opinion
AI agents restructure media buying in favour of independent agencies, says Converge’s CEO
Manchester, the driving force behind the Northern powerhouse, is a standard bearer for the bustling independent media agency scene in the UK. Nowhere was this clearer than in a panel at ‘Future of Media Manchester’ that celebrated the collaboration and entrepreneurial talent in a region that now accounts for a quarter of all UK agencies.
However, as Tom Salmon from Agency by Agency revealed in his presentation, AI remains conspicuously absent from the priority lists of independent agencies. This is a missed opportunity. AI gives indies a competitive edge that can only supercharge their growth.
AI Agents in particular offer independent-minded media agencies an economical way to close the execution and effectiveness gap with their network peers.
Automated buying with built-in dedicated AI agents handling brand safety (e.g., page- and frame-level classification for hate speech, misinformation and suitability by topic), audience targeting (propensity models that blend first-party cohorts with contextual and real-time behavioural signals), and media quality (predicting viewability, invalid traffic and attention uplift before bid) enables nimble agencies to leverage their creativity and flexibility without being held back by a lack of tech investment and know-how.
Media buying needs more than bolt-ons; it needs to be rebuilt
Much of what is marketed as ‘agentic AI’ today is a user interface bolt-on rather than the innovation that digital advertising needs. The real step-change occurs when agents are integrated into the data pipeline and decision loop, allowing advertisers to rebuild buying logic rather than automate yesterday’s processes. That’s the opportunity: not a new interface, but a new architecture.
Digital advertising is in dire need of such disruption. AI agents (defined hereafter as trainable models designed to automate specific functions) provide the opportunity to go back to the drawing board on digital advertising’s core purpose, which is, to quote a vintage 1999 DoubleClick slogan, “Right advertisement to the right person at the right time.”
A quarter century later, and we’re still chasing that dream. In the meantime, digital advertising’s supply chain has become a labyrinth of taped-together hacks and solutions to problems that were caused by other solutions. For agencies, they either need dedicated technologists to make sense of the mess or to hire services from the crowded market of specialist intermediaries.
Limited in their ability to afford either the in-house or outsourced expertise, many independent agencies have been largely relegated to Google and Meta resellers; a frustrating position to be in when you’d rather be delivering ambitious, better-achieving campaigns rather than following the path of least resistance.
What if AI agents didn’t just introduce efficiencies on top of the existing supply chain, but rewrote it from scratch? The result would be a levelled playing field, where access to advanced media buying capabilities is decentralised and democratised, with independent agencies deploying bespoke agents to work for them on a per-campaign basis.
For instance, the unique capabilities of AI agents can be deployed to solve targeting and recruitment problems that couldn’t be done using standard industry means. In an almost instantaneous process, agents, alongside skilled workers, can define the demographics and interests of the audience using a variety of robust, static panel data. Then, agents can combine these pointers with audience behaviours, or signals, through live on-page streaming data.
This results in consistently reaching hard-to-reach and shape-shifting audiences while increasing the level and quality of response that the advertiser needed to justify a new approach.
The radical simplicity of AI agents
The best way to approach deploying an AI agent is to think, if you had the time, how would you do the task yourself?
If it were brand safety, for instance, you would look at the media to determine if there was anything inappropriate for the campaign in question. This determination could only be made after gaining the necessary experience in media literacy and cultural norms and would be applied to one piece of media at a time.
An AI agent can be trained for that same task, if provided with the right labelled dataset and evaluation harness, crunching the years of experience needed for a human to make such decisions down to a few [hours/days/weeks] of training.
The heavy lift is curating and tagging representative examples (including edge cases), establishing ground truth, and monitoring drift. With that in place, training and fine-tuning can be rapid; without it, the model is guesswork.
This knowledge is used to build a probabilistic model that can automatically make brand safety judgements across [hundreds/thousands] of potential media placements, all in the 200 milliseconds it takes for a programmatic transaction to complete.
You can apply this same process to the various components of media buying. Is the media environment high-quality? Train an agent to predict it. Is my target audience likely to see the advert? Same again. Do you know if this placement will trigger the outcome my client wants? You know the drill.
Any aspect of the media buying process that is measurable, repeatable, and supported by high-quality data is open to being automated by an AI agent, provided sufficient engineering effort is invested. Have multiple of these agents working in tandem with a central agent trained to orchestrate their various predictions and feed them into the bidstream, and the cumbersome tech stack that an agency would usually have to juggle to make a buying decision is narrowed down to a single “point of contact”.
Governance, guardrails, and accountability
Whenever AI is discussed, it’s necessary to consider governance and guardrails. Media-buying agents are not safety-critical, but they do influence who sees what and at what price.
Under the EU AI Act, most such systems will be categorised as lower-risk, with transparency and accountability duties, unless they engage in prohibited practices such as manipulative techniques or exploiting vulnerabilities. As such, when building these agents, we should treat governance as proportional, not perfunctory.
Purpose-built AI agents lack the ethical and legal quandaries of generalised large language models fed on copyrighted materials scraped rapaciously from the entire internet. Nor do they carry the thorny questions around creativity and disclosure presented by generative AI. AI agents present a more efficient way of automating a supply chain that was already largely automated.
Beyond GDPR considerations, what is important to note in terms of accountability is that AI models are probabilistic: they produce the most likely output given the inputs and training.
Errors vary by task and design; generative models can fabricate (‘hallucinate’) when prompted beyond their grounding, whereas discriminative task models misclassify under drift or poor data. Guardrails, constrained prompting, retrieval/grounding, and human-in-the-loop substantially reduce these risks when it matters.
In short, AI agents restructure media buying in favour of independent agencies.
By stripping inefficiencies and simplifying essential functions, they’d free up Manchester’s independent agencies to trade on their creative strengths rather than be constrained by limited resources or unwieldy tech stacks.
This creates a game-changing opportunity for them not only to keep pace with larger network competitors but also to leapfrog them. The potential is vast, and the future belongs to those who seize it.
Ian Maxwell is the CEO of Converge