Worked example

Accounts payable: from invoice backlog to a governed AI workflow.

Follow one bounded business question from current work through human–AI allocation, deterministic economics, implementation requirements and outcome observation.

All numbers are illustrative assumptions used to demonstrate the method. They are not customer results or performance claims.

Illustrative example — not a customer or performance claim

Decision in scope

How should invoice intake, matching and exception handling change while people retain supplier, exception and payment authority?

1 · Understand the baseline

Annual volume

12,000 invoices

illustrative assumption

Handling time

18 min / invoice

illustrative assumption

Loaded labour cost

A$55 / hour

illustrative assumption

Calculated baseline

3,600 hours · A$198,000

volume × time × cost

2 · Decide the work allocation

AI-led · 70%

Routine intake and three-way matching

AI extracts, validates and matches approved records. The scenario retains an average of 4 human minutes per invoice for sampling, oversight and routed exceptions.

People work with AI · 20%

Exception research

AI assembles evidence and suggests the relevant policy; an AP specialist resolves the case.

Human-accountable · 10%

Supplier, policy and payment decisions

Authorised people decide material exceptions and release payment. AI receives no approval authority.

3 · Calculate the scenario

Baseline: 12,000 × 18 min = 3,600 hours

Designed human touch: (8,400 × 4 human min) + (2,400 × 10 human min) + (1,200 × 24 human min) = 1,440 hours

Gross human-capacity difference: 3,600 − 1,440 = 2,160 hours

Illustrative gross annual human-capacity value: 2,160 × A$55 = A$118,800

This is not ROI or realised savings. AI runtime is not counted as human touch time; implementation cost, software and inference run cost, ramp-up, demand, risk and adoption must be added before an authorised investment decision.

4 · Deliver the implementation blueprint

  • Invoice intake, matching and exception workflow
  • AP specialist, approver and AI work allocation
  • ERP, supplier and document-data requirements
  • Payment authority, thresholds and exception controls
  • Acceptance tests, rollout sequence and named owners
  • Adoption, handling-time, exception and benefit instrumentation

5 · Measure what actually happens

  • Share of eligible invoices using the new workflow
  • Human minutes and elapsed cycle time by work type
  • Exception, rework and policy-breach rates
  • Capacity released versus the approved scenario
  • Evidence review before any benefit or causal claim

Discuss your challenge

Use the example to challenge the method.

Then bring a real business decision and replace every illustrative input with authorised evidence from your organisation.