Which signs show a set-up workflow is worth using for agent training or evaluation?

A set-up workflow gives useful signal when it has most of the traits below, because each one creates a place where an agent can be right or wrong in a way you can measure.

- **Two or more systems must agree.** The customer, account and service records need to line up, not just exist. - **An existing-record check is required.** Duplicates, reopened records and near-matches force a judgement call. - **Policy changes the path.** Product, region, customer type or risk rating alters required steps or fields. - **Inputs are sometimes incomplete.** The approval omits something a target system demands, and the correct move may be to ask or pause. - **Writes depend on order.** A later record needs an identifier from an earlier one, so partial failure leaves a recoverable or unrecoverable state. - **Someone else needs to know.** A handover note, task or notification is part of the definition of done. - **Success can be checked from state.** You can query the final records and compare them with an expected outcome, instead of judging prose.

If a workflow has only one system and no branching, a simple form-fill test will probably do. If it has four or five of these traits, a generated task pack and simulated environment are likely to be worth scoping in a pilot.