How should a team choose the boundaries of its first complaint pilot?

Scope the first pilot around a decision your agent needs to make, such as whether a completed investigation should be referred to a reviewer. Define what information is available at that point, which actions the agent may take and what outcome a reviewer can assess.

Include enough surrounding process to make the decision meaningful. The case may need an acknowledgement history, investigation findings and an applicable policy. It does not automatically need every upstream intake channel or every downstream remedy system. Those boundaries should be explicit so excluded behaviour is not mistaken for demonstrated capability.

Choose the product form according to the uncertainty you need to resolve. A dataset can support representation and training work. A task pack can test decisions against supplied records. A simulated environment can test how the agent discovers information and changes state through tools.

A useful first step is to write a short pilot brief describing the decision, permitted actions, exclusions and acceptance evidence. MirrorFirm could use that brief to shape a generated sample. Review the sample against the intended agent role before expanding the workflow or adding more business variation.