9 Nov 2026 · 3 min read

What Approvals, Data and Integrations Are Needed to Start a Scoped Marketplace Pilot?

Why "pilot" needs a definition before it needs a checklist

A pilot that is not scoped tends to drift into a full deployment decision before anyone agreed to make one. The first piece of groundwork is not technical, it is defining what the pilot is actually meant to answer: a specific question such as whether agent-negotiated deals hold up for one inventory type, or whether a buy-side and sell-side agent can transact without constant human intervention for a bounded set of deals. Everything else follows from that scope being narrow and explicit.

The approval that matters most: who can say yes to a live deal

Before any technical work starts, a pilot needs a named person with the authority to approve deals at the agreed scope, not a committee or a vague "the team will review." An agentic marketplace pilot moves at a different pace to a traditional negotiation, and an approval process that assumes human review happens on the same cadence as email will become the bottleneck rather than the safeguard. The approval point should be specific: which deals need sign-off before they close, and who is actually available to give it.

The data a pilot actually needs, and the data it doesn't

A scoped pilot needs enough data for the agent to make a defensible decision within the agreed scope, not a full export of historic campaign performance. For a buy-side pilot, that typically means the inventory or audience parameters relevant to the specific test, current budget limits, and any exclusion rules that must be respected. For a sell-side pilot, it means the inventory being offered and the floor terms a selling agent is permitted to accept. Scoping the data narrowly to the pilot's actual question, rather than granting broad access up front, keeps the pilot's risk proportional to what it is actually testing.

Integrations: the minimum viable connection, not the full stack

A pilot does not need every system connected on day one. It needs the specific integration that lets the agent see the inventory or budget data relevant to its scope, and a way for a human to review and approve deals within that scope. A full integration into every adjacent system, order management, finance, reporting, can wait until the pilot has answered its actual question. Attempting the full integration before running the pilot is one of the most common ways a scoped test turns into a stalled deployment project.

What a neutral marketplace needs from its side

On the marketplace side, a scoped pilot runs on a DealSheet for every deal struck during the pilot: a record of the agreed terms, timestamped and held independently by both the buy-side and sell-side agent, so either party can review exactly what was agreed without relying on the other side's account. That record is what lets a pilot's results be reviewed honestly afterwards, deal by deal, rather than summarised from memory.

Keeping the pilot a pilot

The discipline that keeps a pilot useful is resisting the urge to expand its scope mid-test. A pilot that starts narrow, with a named approver, minimum necessary data access, and the smallest integration that answers the actual question, produces a result a team can trust. One that grows broader as it runs usually produces neither a clear answer nor a controlled test.

All articles

Speak to the team

Request a demo

You don't need to plug in your agent, the demo walks you through the steps.