Skip to course content
Free agentic AI fundamentals course

Introduction to Agentic AI and Workflow Automation

Unit 09.01: Mapping the approvals that already exist

The automation has to match the approval structure the organisation already has. Simplifying it is not your decision to make.

Bands, latency, and how many approvers

Three amount bands with different approvers, different latencies and different approver counts.

The table below lays them out.

case                   approver           latency    approvers
credit under 100       team lead          same day   1
credit 100-1000        finance manager    1-2 days   1
credit over 1000       finance director   2-5 days   2

Two consequences follow directly. A pause measured in days means the run has to
survive being put down -- durable state, not a connection held open.

And "2 approvers" means the state records both individually, or a second
approval can be satisfied by the same person clicking twice. That check is
never written until an auditor asks for it.

The latency column changes the architecture. A pause of two to five days means the run cannot hold anything open - durable state, from Module 4, rather than a process waiting.

"2 approvers" has its own consequence: the state records both individually, or the same person clicking twice satisfies the requirement. That check is never written until an auditor asks for it.

The mistake this prevents

The mistake is building one approval step and mapping every case onto it. The bands exist because the organisation decided different amounts warrant different scrutiny, and collapsing them either over-escalates small items or under-escalates large ones.

Takeaway

Model the approval chain as it exists - approver per band, latency in days, approvers recorded individually. Long pauses mean durable state.