A machine-readable deal is a deal structured so that AI agents can read, interpret, and act on the terms without a human translating at each step. In agent-to-agent advertising, this is what makes autonomous negotiation possible. The deal is not a document a human reads and interprets; it is a structured data object that two agent systems can exchange, verify, and record without ambiguity.

What is a machine-readable deal?

A machine-readable deal is a structured representation of agreed terms in a format that software systems can parse and act on directly. Unlike a contract in natural language or a deal summary in a spreadsheet, a machine-readable deal has each term encoded in a defined field with a defined data type. The buying agent and the selling agent can both read the same deal record, confirm the terms match what was negotiated, and use the record as the basis for any downstream action.

The key property is unambiguity. A human-readable deal leaves room for interpretation. A machine-readable deal does not: each field has a specific meaning, a specific format, and a specific value that both systems have agreed to.

What fields does a machine-readable media deal include?

A machine-readable media deal typically includes: the identity of the buyer and seller, the inventory being traded (site, format, and placement), the price and pricing model, the campaign flight dates, any audience or contextual targeting parameters, brand safety requirements, and the approval status of the deal. It may also include the negotiation history: the sequence of offers and counter-offers that led to the agreed terms.

Each of these fields is encoded in a format the agent can read directly. There is no need for a human to extract the price from a contract paragraph or confirm the flight dates from an email thread. The deal record contains all relevant terms in a form both systems can verify.

How does the deal format change the negotiation process?

When the deal format is machine-readable, negotiation can happen between agent systems directly. The buying agent proposes a deal record with its initial terms. The selling agent reads the record, checks whether the terms fall within its acceptable range, and either accepts, declines, or returns a counter-proposal with modified field values. The process repeats until both sides accept the same deal record or the negotiation fails.

This is structurally different from negotiation by email or by API request. Both systems are working against the same data structure, so there is no translation step and no risk of one party's understanding of the terms diverging from the other's.

Who holds the deal record?

In a well-structured agent-to-agent deal, both parties hold the same deal record. The buyer has a copy and the seller has a copy, and both copies are identical. Neither party depends on the marketplace to reconstruct what was agreed: the record is held directly.

This bilateral ownership is significant. When a deal is held only by the exchange, the exchange is the authority on what was agreed. When both buyer and seller hold the deal record independently, they can compare records and identify any discrepancy without relying on a third party to arbitrate.

What happens when the deal terms are not met?

When a campaign runs against a machine-readable deal, the delivered campaign can be checked against the agreed terms field by field. If the placement delivered does not match the placement agreed, or if the volume falls short of the committed figure, the shortfall is measurable against the deal record rather than against a platform's reporting output.

This makes dispute resolution more tractable. The question is not what the platform says happened but whether the delivered campaign matches the terms both parties signed. The deal record is the reference.

How does this connect to audit and accountability?

Machine-readable deals are the foundation of auditable agentic advertising. If the deal terms exist only in natural language or are reconstructed after the fact from platform logs, audit is limited to what the platform chooses to report. When the agreed terms are encoded in a structured record held by both parties, any third-party auditor can compare the record against the delivered campaign without depending on the marketplace's co-operation.

This is the accountability mechanism that makes agent-negotiated advertising credible for buyers who cannot monitor every agent decision in real time. The deal record is the evidence that the agent acted within the agreed terms.

The DealSheet is Alkimi's implementation of a bilaterally-owned, machine-readable deal record. Buyer and seller both hold the record; it is not owned by the marketplace. This means either party can verify what was agreed without depending on Alkimi to confirm it.