How does escalation work in a real business, and why would an AI team care?

**What happens when a case exceeds a reviewer's authority?**

The reviewer finds that the amount, risk rating or exception is beyond what their role may decide and passes it up with a reason. The senior reviewer sees the case, the first reviewer's notes and the policy basis, then approves, amends, declines or returns it for more information.

**Is it always upward to one person?**

No. Some organisations route by amount band, others by product or region, and some require two approvers above a certain level. A dataset can include these variants so an agent cannot memorise a single path.

**Do human reviewers always escalate correctly?**

Not always. Real processes contain cases escalated without the needed evidence, cases that bounce back, and cases decided slightly outside authority and corrected later. A generated task pack can represent these as scenarios with labelled outcomes.

**Why would an AI team care?**

The behaviour you want is conditional. An agent that approves whatever it technically can is not safe in an approvals workflow, and an agent that escalates everything adds no value. This workflow supplies cases on both sides of the line.

**Is the data real?**

No. MirrorFirm output is generated. It is informed by how businesses operate, but you should inspect any release for grounding and disclosure risk before using it.