What is a useful first pilot for an interactive referral environment?

Start with a specific boundary: a utility service adviser recognises payment difficulty and transfers a valid referral to a specialist queue. Specify the incoming evidence, permitted actions and condition that establishes successful transfer. This gives the pilot a coherent beginning and end.

Keep downstream assessment outside that first slice unless it is essential to the buyer's intended test. An environment that covers recognition, referral, assessment and long-term account management needs several distinct policies and evaluators. Combining them too early makes failures harder to interpret.

Ask MirrorFirm for a generated sample that exercises the chosen boundary through ordinary progress and a meaningful exception. Review the records and action outcomes together, so the sample demonstrates executable behaviour rather than only credible prose.

Agree acceptance criteria before expanding scope. These can cover schema validity, policy consistency, recoverable errors and the ability to reproduce a run from its starting state.

The first useful deliverable is a reviewed environment specification plus a runnable sample against the buyer's interface. That establishes what needs further generation, what requires simulator work and which unresolved business rules need an explicit decision.