What does intake look like in a real business, and why does it suit agent evaluation?

Intake is the point where something arrives, such as a loan application, an insurance claim notice, a utility connection request or a patient referral, and the organisation decides what it is, whether it is complete, and who handles it next.

In a real business this is rarely one step. A case arrives through several channels: a web form, an email, a broker portal, a phone note. Someone captures the fields, matches the applicant to an existing customer record, checks mandatory items against a policy, chases missing documents, and routes the case to a queue based on product, value, risk or region.

For evaluation teams the workflow is useful because the ground truth is checkable. Completeness rules are explicit, routing tables are finite, and errors have visible downstream cost. A case sent to the wrong queue stalls, and a case marked complete when it is not causes rework in another team.

It also contains the awkward parts that simple benchmarks skip:

- partial submissions and later amendments - duplicate or near-duplicate submissions - documents that contradict form fields - policy exceptions that need a named approver - handovers where context is lost

A generated dataset or environment for this workflow can encode those conditions with known answers, so an agent's capture, check and routing decisions are scored against stated rules. The material is generated, so buyers should still inspect how closely it is grounded in realistic practice.