Supply Intelligence
A supply descriptor is a structured record of what inventory is. But describing something accurately is not the same as understanding its value for a specific purpose.
A wine list describes each bottle: region, grape variety, vintage, producer. But a sommelier adds intelligence to that description — drawing on knowledge of the diner's preferences, the occasion, the food being served, and the house's inventory history to recommend the right match. The supply descriptor is the wine list. Supply intelligence is the sommelier.
Supply intelligence is the process — and the layer of infrastructure — that transforms raw supply descriptors into scored, ranked, decision-ready signals that a buyer agent can act on with confidence.
What Description Misses
A supply descriptor tells a buyer agent what a piece of inventory is. What it does not tell the buyer agent, on its own, is:
- How well does this inventory match a specific campaign's intent, across all the relevant dimensions simultaneously?
- How reliable are the descriptor's claims, given the publisher and the system that produced it?
- How does this opportunity compare to the other opportunities available right now?
These are intelligence questions, not description questions. Answering them requires more than reading a descriptor — it requires evaluating a descriptor in context, applying scoring logic, and producing a signal that a buyer agent can use to rank and filter options quickly.
The Confidence Score
The primary output of supply intelligence evaluation is a confidence score: a normalized signal indicating how well a specific supply opportunity matches a specific demand signal.
Confidence scoring is not a single number that ranks all inventory on a universal scale. It is a function of both the supply and the demand — the same piece of inventory may receive a high confidence score for one campaign brief and a low one for another, depending on the alignment between them.
The inputs to a confidence score typically include:
Semantic alignment. How closely does the content's topic coverage match the campaign's intent? A buyer brief for professional kitchen equipment and a page about restaurant kitchen design have high semantic alignment; a page about consumer recipes has lower alignment.
Brand suitability match. Does the inventory's suitability declaration meet the campaign's thresholds? A campaign with strict brand safety requirements will score inventory with ambiguous suitability declarations lower, regardless of topical relevance.
Freshness fit. Is the content's age appropriate for the campaign? A campaign running time-sensitive news advertising will discount older content; an evergreen information campaign will not.
Provenance completeness. How fully attested are the descriptor's claims? High-provenance descriptors — where the claims are backed by verifiable audit trails — earn higher scores than descriptors with minimal attestation.
In practice
A buyer agent running a campaign for a cybersecurity software brand evaluates two supply opportunities. The first is a recent article about enterprise security vulnerabilities, on a recognized technology publisher, with a fully attested supply descriptor. The second is an older article on a general news site covering the same topic, with an auto-generated descriptor and no external verification. Both are topically relevant. The confidence scoring system gives the first opportunity a significantly higher score — not because of the topic, but because of freshness, publisher quality, and provenance completeness. The buyer agent acts on that ranking.
Relationships Over Time
Supply intelligence does not evaluate each descriptor in isolation. Over time, a supply intelligence system accumulates a structured model of relationships between supply sources, content types, topics, and historical context.
This relational model allows a system to reason about patterns that go beyond any single descriptor. Has this publisher's supply historically attracted demand aligned with a given intent category? Does this topic area tend to draw high-quality campaigns, or primarily low-value remnant demand? Are there publishers whose descriptors are highly consistent and accurate, versus others whose claims diverge frequently from observed outcomes?
The performance signals referenced here are supply-level — aggregated patterns about how inventory is described, matched, and transacted — not user-level behavioral profiles. They describe what happened at the supply and campaign level, not who was visiting.
None of these questions can be answered from a single descriptor. They require relationships built up over time — which is why supply intelligence is not just a scoring function, but an ongoing system that learns and updates as more inventory and outcomes are observed.
Intent Resolution
The demand side of supply intelligence is intent resolution: the process of translating a campaign brief — expressed in natural language or structured fields — into a form that can be matched against supply descriptors.
A campaign brief that says "professional development content for finance industry professionals" needs to be resolved into specific topic categories, brand suitability thresholds, and publisher preference signals before a buyer agent can evaluate supply descriptors against it. Intent resolution is that translation.
The output of intent resolution is not a keyword list. It is a structured demand signal: a typed representation of what the campaign is looking for, expressed in the same vocabulary that supply descriptors use. This shared vocabulary — aligned by protocol standards like AdCP — is what makes it possible for buyer agents and supply intelligence systems to communicate without custom integration for every pairing.
Why Intelligence Matters
The difference between a supply descriptor alone and supply descriptor plus intelligence is the difference between a list and a ranking.
A buyer agent could, in theory, read every supply descriptor available and evaluate each one from scratch. In practice, the available inventory pool is enormous, the time available for evaluation is measured in milliseconds, and the quality of that evaluation needs to be high enough to justify autonomous decision-making.
Supply intelligence compresses this problem. By pre-scoring descriptors against known demand patterns, maintaining a relational model that encodes historical context, and resolving campaign briefs into structured demand signals, the intelligence layer reduces the buyer agent's job to acting on ranked, pre-evaluated options rather than performing evaluation from scratch on every impression.
Companies building supply intelligence systems — including Modulr and others — are, in effect, building the reasoning infrastructure that makes buyer agents useful at scale. The next unit covers the provenance layer that makes this reasoning trustworthy.