Suggested internal effort: 4–10 person-days; external response time unknown. Research direct purchasers, representatives, collective routes and distribution channels. Do not send raw archive material to everyone found in a search.
For each prospect, record the original announcement or current requirement, its date, modality, use, buying entity, procurement route and evidence limitations. Ask a genuine business contact whether the requirement is still active. Use permissioned introductions and official business enquiry channels; avoid inventing contact details.
The first discussion should establish the problem the buyer is solving, a suitable cohort, technical requirements, available budget or procurement stage, timing, and the basis for a restricted evaluation. Record a refusal or unanswered enquiry honestly. “Confidential pipeline” is not a substitute for enough evidence to decide whether to invest.
Set your own advance criterion. For example, an owner might require one documented requirement and a named buyer-side technical reviewer before spending on a pilot. That is a proposed gate, not an industry standard and not proof a contract will follow.
Completion condition: a source-backed requirement exists for the pilot, or the sponsor consciously authorises a small research budget to test an unproven hypothesis. If only general AI interest has been found, continue research or stop; do not describe it as qualified demand.