Supply path optimisation has been a topic of conversation in programmatic advertising for several years. The term has genuine meaning, but it is often used as a proxy for a range of different initiatives, some of which are genuinely about transparency and some of which are about cost reduction dressed up as transparency. Before asking what SPO means in an agentic market, it is worth being clear about what it means in any market.

What is supply path optimisation?

Supply path optimisation is the practice of reducing the number of intermediaries between an advertiser and a publisher in order to improve three things: transparency (knowing where money goes and what decisions are made along the way), cost efficiency (reducing the fees paid to intermediaries who do not add value proportionate to their take), and data fidelity (ensuring that the audience and inventory data used to make buying decisions is not degraded or altered in transit).

The problem SPO addresses is structural. The programmatic supply chain grew by accretion. Each intermediary that entered the chain provided a service or solved a problem at the time of entry. Over time, the chain became long, opaque, and expensive, and many of the intermediaries in it were providing services that had become redundant or were impossible for buyers to verify. SPO is the discipline of asking, for each step in the chain, whether that step adds value that justifies its cost and its opacity.

Done well, SPO is a buyer-side governance practice. Done badly, it is a negotiating tool that shifts fees without actually reducing complexity or improving transparency.

Why was SPO already a pressure on the industry before AI buyers existed?

The commercial case for SPO does not depend on AI. It existed before AI buying was a viable option and it would exist if AI buying never arrived. The intermediary layer in programmatic advertising is genuinely expensive and genuinely opaque, and buyers who do the work to understand what they are paying for consistently find that a significant portion of their media spend is absorbed by fees and margins that cannot be directly mapped to outcomes.

The industry responses to this have been varied. Some buyers have invested in direct publisher relationships and reduced reliance on the open exchange. Others have worked with a smaller number of supply-side partners and negotiated greater transparency as a condition of the relationship. The Association of National Advertisers and equivalent bodies have published research showing that large portions of programmatic budgets do not reach publishers in any form that translates to working media. None of this is new, and none of it required AI to be the buying agent.

What AI buying does is change the stakes. Not the underlying problem. But the consequences of leaving the problem unresolved.

How does an agentic market change the SPO problem?

An AI buyer working from a brief needs to evaluate inventory quality and deal terms at speed, across many sellers, without a human reviewing each decision. The brief specifies what the buyer wants: audience, channel, timing, quality floor, budget. The AI then negotiates to assemble a portfolio that satisfies those specifications.

When the supply path between the AI buyer and the publisher is long, the information available at each negotiation step is degraded. The AI cannot directly verify inventory quality claims made through three intermediary layers. It cannot confirm that the audience segment it is buying is the audience segment it thinks it is buying, because the data has passed through systems it cannot inspect. It cannot verify the provenance of the inventory.

A human buyer encountering the same problem can apply judgement, experience, and scepticism that is not encoded in a brief. They can call a publisher directly, ask for verification, or rely on a relationship that has been built over time. An AI agent working at machine speed through a defined interface cannot do those things in the same way. The opacity that was always a problem becomes a specific liability when the agent making decisions cannot see through it.

What is the logical destination of agent-to-agent trading in terms of supply path?

If a buy-side agent and a sell-side agent can negotiate directly, the intermediary layer between them serves a different purpose than it does in the current model. In the current model, intermediaries provide services that neither buyer nor seller is resourced to provide directly: auction mechanics, audience data enrichment, brand safety filtering, fraud detection. Some of these services remain necessary regardless of whether the parties are humans or agents.

But the current intermediary layer also exists, in significant part, because buyers and sellers could not communicate directly at the speed and scale required for real-time transactions. When agents can negotiate directly, at machine speed, that specific justification for the intermediary layer weakens. The remaining question is which services in the current chain provide genuine value that agents cannot replicate or replace, and which are artefacts of a model built for human-speed transactions.

The answer will not be the same for every service in the chain, and the transition will not be immediate. But the direction is clear: agent-to-agent trading applies sustained pressure to intermediaries who cannot explain their value in terms that make sense when the parties at either end are AI systems rather than humans.

What does a deal record add to the SPO picture?

In a governed agentic buying model, the AI commits to deals and produces a record of what was agreed, under what terms, and why. That record is a new source of supply chain data. It shows not just what was bought and at what price, but which sellers were approached, which terms were negotiated, and where the final portfolio deviated from the opening brief.

For SPO purposes, this is significant. The deal record makes the supply path legible at the transaction level in a way that aggregate reporting cannot. If an agent consistently reaches a particular publisher through three intermediary steps when a more direct path exists, that pattern is visible in the record. If the terms negotiated through one path are systematically worse than through another, that is also visible.

SPO decisions have always required data to make well. The deal record produced by an agentic buying system is more granular and more actionable than most of the supply chain data that buyers currently have access to. The transparency that SPO has always depended on becomes a structural output of the buying process rather than something that must be reconstructed after the fact from aggregate reports.

Why does agentic buying make transparency load-bearing rather than good practice?

SPO has long been described as good practice: a sensible approach to procurement that a well-run buying organisation should pursue. In an agentic market, it becomes something closer to a prerequisite. An AI agent that cannot verify the quality of inventory it is committing to, because the supply path is too opaque, will make systematically worse buying decisions than one operating on a transparent path. The performance gap between a well-governed agentic buyer and a poorly-governed one is larger when the supply path is opaque, because the poorly-governed buyer's errors compound.

Transparency was always about trust between the parties in a transaction. In an agentic market, it is also about the quality of the information available to a machine that must make consequential decisions without stopping to ask a human. Those are different requirements, and the second is stricter than the first. Supply path optimisation was always a good idea. In a market where agents are doing the buying, it is the foundation on which the whole model rests.