Supply path optimisation changes from a periodic human exercise into a continuous automated one when an AI agent runs it, but the underlying problem it addresses does not change at all. The waste, the hidden fees, and the low-quality inventory that SPO exists to cut are still there, and an agent optimising the path can attack them faster and more consistently than quarterly human reviews ever did. What does not change, and what agents can quietly make worse, is the transparency the whole exercise depends on: an agent choosing a path is only as good as the data it can see and trust about that path.
TL;DR. SPO is the practice of choosing the most direct, transparent route for a bid to reach inventory, cutting unnecessary intermediaries and hidden fees. Agents change the execution: they can evaluate and reselect supply paths continuously, factoring fee visibility and quality signals into every decision rather than in occasional human passes. What stays the same is the reason SPO exists, the structural waste documented across the programmatic supply chain, and the dependence on data the buyer can actually verify. The new risk is that an agent optimising a path it cannot fully see may route efficiently toward the wrong place, which turns the old transparency problem from a reporting inconvenience into a decision input.
What is supply path optimisation, and why does it exist?
Supply path optimisation is the practice of deliberately choosing how a bid reaches inventory, rather than accepting whatever route the pipeline serves. The same impression can often be bought through several paths, each with different intermediaries taking different fees, and SPO is the discipline of favouring the paths that are more direct, more transparent, and less wasteful.
It exists because the programmatic supply chain leaks value at a scale the industry has measured and struggled to fix. The Association of National Advertisers, in its programmatic transparency work, found that of every dollar entering a demand-side platform, only around a third reached a consumer, with a large share lost to intermediary fees and to low-quality or unmeasurable inventory. The waste has not gone away since. The ANA's more recent benchmarking put the unrealised value in the tens of billions and noted that even as some specific problems shrank, overall inefficiency rose, with the composition shifting from obvious fraud toward structural inefficiency: indirect supply paths that add fees without adding value. SPO is the buyer's tool against exactly that structural leakage.
What changes when an agent runs SPO?
The cadence changes first, and it is the most consequential change. Human SPO happens in periodic reviews: a team analyses supply paths, cuts the wasteful ones, and reconfigures buying, then revisits it weeks or months later. An agent evaluates paths continuously, as part of every buying decision, which means a path that degrades, a route that starts adding fees or serving worse inventory, can be dropped as it degrades rather than at the next quarterly review. Continuous beats periodic when the thing being optimised is itself always shifting.
The scope changes second. Human SPO tends to operate at the level of configuration: which exchanges to prefer, which paths to exclude, set as standing rules. An agent can make the choice at the level of the individual decision, weighing the fee visibility and quality signals of each available path against the specific goal of the campaign at that moment. It is the difference between a policy and a judgement made fresh each time.
The integration with the rest of the buy changes third. For a human, SPO is a somewhat separate discipline, run by specialists and connected loosely to campaign optimisation. For an agent, path selection is just one of the levers it optimises alongside targeting, budget, and bidding, and it can reason about the relationships between them, recognising, for instance, that the reason a segment underperforms is the path it is being bought through rather than the audience itself. Path selection stops being a separate exercise and becomes part of one continuous optimisation.
What does not change?
The problem does not change. The waste, the hidden fees, and the low-quality inventory that SPO exists to cut are structural features of the supply chain, and an agent does not remove them by being fast. It navigates them better, which is valuable, but the leakage is still there to be navigated, and an agent pointed at the wrong objective will navigate it no better than a human would.
The dependence on data does not change, and this is the crucial continuity. SPO has always been limited by what the buyer can see. The ANA's own conclusion was that the inefficiency persists partly because of a lack of visibility, and that access to detailed, log-level data was what allowed buyers to find where value was hiding. That dependence is not softened by automation. An agent choosing a path needs accurate data about that path's fees and quality, and where that data is missing or unreliable, the agent is optimising in the dark, just faster than a human would be.
The need for verification does not change either. A human running SPO could at least apply judgement and skepticism to a suspicious path. An agent applies whatever its data tells it, so the burden of ensuring that data is trustworthy moves squarely onto the infrastructure. What SPO required of the buyer, insist on transparency into the supply chain, it now requires of the systems feeding the agent.
What new risk do agents introduce to SPO?
The new risk is efficient routing toward the wrong destination. A human doing SPO badly does it slowly, and the errors are visible and correctable at the next review. An agent doing SPO on incomplete or unreliable path data does it continuously and at scale, confidently selecting paths that look optimal on the data it has while being wrong on the data it lacks. The old transparency problem was a reporting inconvenience: you found out about the waste after the fact. The agentic version is a decision input: the agent acts on the gap in real time, so opacity now corrupts the decision itself rather than just the report about it.
This is why the transparency problem and the agentic transition are more entangled than they first appear. The same opacity that let fees hide in a supply chain, the thing SPO was invented to fight, now degrades the quality of an agent's path decisions directly. An agent cannot optimise around a fee it cannot see or a quality problem its data does not capture. Automation raises the stakes on transparency rather than lowering them, because it removes the human pause in which a person might have noticed something was off.
There is a further wrinkle specific to agentic buying. When agents on both sides of a deal transact, the record of what was actually bought and delivered can diverge between them, which means even the after-the-fact data used to evaluate a path's performance may not be something both parties agree on. As one on-record investigation into agentic buying put it, an agent asserting a transaction occurred is a claim rather than a receipt, and a deterministic, auditable record is what separates the two. Published simulation research into agentic reconciliation found that two agents which had just agreed a deal recorded its terms differently in the large majority of cases, with the divergence compounding over the campaign. That work models the specification architecture rather than any live system, but it points to a real complication for agentic SPO: optimising a supply path assumes reliable data about how that path performed, and if the two sides do not agree on the performance record, the optimisation is being tuned against a contested number.
What should a buyer do about SPO in an agentic setup?
Treat supply-path data as a first-class requirement, not a reporting afterthought, because in an agentic setup it is a live decision input. Before letting an agent optimise paths, establish what data it has about fees and inventory quality on each path, and how reliable that data is, because the quality of every path decision is capped by it. An agent with rich, trustworthy path data will optimise SPO better than any human process. An agent with thin or unreliable data will make fast, confident, wrong choices.
Ask the same transparency questions of an agentic platform that SPO taught buyers to ask of the supply chain, now aimed at the agent's inputs. Can the agent see the fees on each path? Can it assess inventory quality on signals the buyer trusts? And can the buyer verify, after the campaign, which paths the agent actually chose and how they performed, against a record both sides of the deal accept? SPO in the age of agents is not a new discipline. It is the same discipline, with the transparency requirement moved from the report to the decision, where getting it wrong now costs more and faster.
This article references the Association of National Advertisers' programmatic transparency research on supply-chain waste and the value of log-level data, and cites published simulation research into agentic deal reconciliation conducted by Alkimi. It also references the IAB Tech Lab's Agentic Advertising Management Protocols (AAMP) work on shared, auditable transaction records. The simulation models the current specification architecture and is not an assessment of any specific production platform.