Name the decision you want to improve
Describe the user workflow and the specific decision or task that needs support. Identify the cost of an incorrect result and where human review belongs. This gives prospective specialists a concrete problem to assess.
Understand the data you can use
Record where data lives, who owns it and what access restrictions apply. Note known quality gaps and whether representative evaluation examples exist. Avoid sharing confidential datasets during early candidate conversations.
Agree what useful means
Choose evaluation criteria alongside the team that will use the system. Include reliability, latency, operating cost and the quality of the user experience. Start with a bounded experiment and make the next investment depend on the evidence.