How should I scope a first escalation pilot with MirrorFirm?

A useful first pilot is narrow and tied to a question you can answer from the results.

**Choose one domain and product area.** For example, small business lending, premium funding, customer hardship write-offs or credit holds. The domain should resemble the process your agent will operate in.

**Fix the roles.** Three or four reviewer grades are usually enough to express meaningful authority differences. Provide your own role names if you have them.

**Pick the triggers.** Decide which of these to include: amount over limit, aggregate exposure, policy exception, risk grade, conflict of interest, reviewer unavailable, return for information.

**State the question.** Examples: does our agent handle aggregate exposure, or does it escalate too often? Naming the question shapes which tasks are included and how they are labelled.

**Decide the form.** A dataset, a task pack with expected outcomes, or a simulated environment with tools and reset each suit different needs. Many teams begin with a task pack and add the environment later.

**Set limits early.** Agree intended use, the number of task variations you want, and how results will be reviewed.

Scope, effort and commercial terms are agreed in discussion, not assumed. The output is generated material, so include a grounding and disclosure review in the pilot plan.