In Part One I Stuart Bailey, co-founder and managing director of AgencyFit wrote about advertisers automating a structure they never fixed. In Part Two, he wrote about publishers solving for a specification no one had published. This closes the loop. Of the three parties in this ecosystem, agencies are the only one that sits in the middle of both relationships at once, and the only one with enough visibility to fix what the other two pieces described.
The Party That Can See Both Sides
An advertiser sees its own agency relationship. A publisher sees its relationship with however many agencies it deals with. Only the agency sees both directions at the same time: the advertiser’s brief, budget and real objective on one side, the publisher’s inventory, product and pricing on the other. That’s not a burden. It’s the only vantage point in the ecosystem from which the whole workflow is visible, and the only position capable of building a bridge between the other two rather than waiting for one to appear.
The five major holding groups have built serious agentic capability on the strength of exactly that position: WPP Open, Publicis CoreAI, Omnicom Omni, Dentsu Connect, Havas Converged.ai. Independents are building the same advantage a different way, often through specialist agentic tools rather than a single in-house platform, and with less distance between the person setting strategy and the person running the system. Different build, same vantage point. The advantage in this piece belongs to the position an agency holds in the workflow, not to its scale.
What Stays Closed
None of this is an argument for giving away the engine. What an agency’s system has learned to weight, its scoring logic, the specific mechanics that turn an input into a ranked recommendation, represents genuine investment and a genuine competitive advantage, whichever kind of system produced it. A publisher or advertiser asking to see inside that logic is asking for something an agency has every right to keep.
That’s a different thing from keeping the ecosystem in the dark about how the system works in shape, if not in substance. The competitive advantage can stay inside the engine. The ecosystem advantage sits in the interfaces around it.
What Can Open Up
I made this argument from the other side in Part Two: publishers cannot build to a specification that’s never been published, so the cost of that silence lands on the publisher. The fix doesn’t require an agency to open its model. It requires something narrower: how each agency plans, the brief language and evaluation criteria a strong submission needs to speak to, the shape of what “good” looks like. That’s already sitting inside agency planning documents and brief templates. Structuring it and making it available is a curation task, not a disclosure of IP.
A publisher who knows what an agency’s planning process is looking for can build a better product, and a better response. Imagine an agency system looking for incremental reach against a specific audience, while a publisher has exactly the right product but has never structured it in a form the system can discover or evaluate. Nobody has made a bad decision. The opportunity has simply been made invisible by the architecture.
An agency that shares that, without exposing the logic that scores it, loses nothing and fixes the exact asymmetry the last piece described.
Linking Up, Not Locking In
The same logic runs the other way, toward advertisers, and it’s narrower there too. An agentic system built well for the agency’s own workflow, but never connected to the advertiser’s own category dynamics, KPIs and campaign history, produces the same failure I described in Part One: two efficient halves that don’t meet. What the system can see is only ever a single brief and a single conversation, unless it’s linked to a standing record of what the advertiser’s business needs and how past campaigns performed, not just briefed on it once and left to infer the rest.
Connecting the agency’s AI to that standing context doesn’t weaken its advantage. It makes the system more useful. Instead of treating every brief as a fresh start, the AI can draw on the advertiser’s objectives, category context, past decisions and campaign performance. The result is a recommendation grounded in the advertiser’s business, not just the brief in front of it.
Build The Bridge
Across all three pieces the same shape keeps appearing. Advertisers who treat a broken workflow as a cost review instead of a redesign. Publishers who chase multiple unpublished specifications instead of building one legible record of their own. Now agencies, sitting at the only point in the triangle that can see both problems at once, facing the same choice in a different form: keep the advantage that’s genuinely theirs, and build bridges out of the parts of the relationship that were never proprietary to begin with.
The industry has spent the best part of two decades building moats, systems that make one party smarter in isolation. Agencies are the party with the standing to build something different, because they’re the only one positioned to see what connecting the other two would require.
Agency AI doesn’t need to be open. But it does need to connect. Where that connection happens, the whole ecosystem gets faster and better at once.
Where it doesn’t, each side keeps optimising its own part, and the gap between them keeps compounding, just as it has in every piece of this series.
