An AP team reports a rising automation rate, yet its inbox remains full and month-end still requires overtime. Both observations can be true. The automated population may contain the easiest invoices, while the remaining cases require more investigation, coordination and judgement.
An automation percentage becomes useful when it is connected to effort, cost and control. Without those measures, it can reward moving transactions through one stage while leaving the difficult work elsewhere in the organisation.
The first step is to ask precisely what the percentage counts.
Define the numerator and denominator
“Automated” could mean an invoice was captured without rekeying, matched successfully, routed without intervention, posted to the ledger or completed without any human touch. These are different achievements.
A rate calculated only across invoices eligible for a particular workflow will also differ from one calculated across every invoice received. Excluding unsupported suppliers, formats or entities may be reasonable for measuring a component, but those cases remain part of the finance operation.
APQC's AP benchmark collection spans cost per invoice, first-time-error-free disbursements and payment cycle time. Its range of measures is a useful reminder that process performance has several dimensions. APQC, accounts payable key benchmarks
For an internal dashboard, state the start event, end event, eligible population and treatment of reopened cases. A colleague should be able to reproduce the percentage from the underlying case records.
A higher rate can coexist with more work
Consider this illustrative monthly operation. In both periods it receives 10,000 invoices. The second period has fewer invoices needing a person, but its remaining cases take longer to resolve.
| Measure | Period A | Period B |
|---|---|---|
| Total invoices | 10,000 | 10,000 |
| Invoices completed without a human touch | 8,000 | 9,000 |
| Automation rate | 80% | 90% |
| Invoices needing human work | 2,000 | 1,000 |
| Average handling time for those invoices | 6 minutes | 15 minutes |
| Total human handling time | 200 hours | 250 hours |
The reported automation rate improves by ten percentage points while human handling time rises by 25%. At an assumed £35 an hour, monthly capacity consumed rises from £7,000 to £8,750.
These are constructed figures to demonstrate the metric, not industry benchmarks or Curia results. A change in case mix, policy or measurement could produce this pattern; the table does not establish a cause.
The lesson is to measure the time and composition of the remaining queue. Reducing its size is helpful only to the extent that the business outcome improves.
Inspect the work around a mismatch
Modern systems can perform substantial matching. Microsoft documents invoice matching against purchase orders, tolerances and product receipts. A failed match may correctly identify a problem requiring investigation. Microsoft Learn, AP invoice matching
The next activity might belong to procurement or a warehouse rather than AP. Somebody must confirm the agreed price, record a receipt, obtain a corrected invoice or approve an exception. If the dashboard measures only the AP user's time, it can miss effort transferred to another department.
Count active handling time wherever the case creates work. Separately measure elapsed time, including waiting. An invoice may require ten minutes of active work spread over five days because nobody owns the next action. Faster data extraction will do little for that delay.
This distinction helps locate the intervention. Long active time suggests an evidence-gathering or processing problem. Long waiting time may call for clearer routing, escalation or upstream accountability.
Use a compact set of operating measures
Start with six complementary views:
- Coverage: the share of all received invoices within the workflow's scope.
- Human effort: total minutes per completed invoice, including correction and work outside AP.
- Quality: first-time-right outcomes, reopened cases and incorrect postings.
- Elapsed time: median and tail resolution times, segmented by exception type.
- Backlog: open cases by age, value, due date and current owner.
- Total cost: technology, service and retained operating effort across the same boundary.
Report performance by meaningful categories. A blended average can conceal an entity with poor receipt discipline or a supplier creating repeated price discrepancies. The categories should lead to an action, rather than produce a more decorative dashboard.
APQC's March 2026 analysis reported AP costs of approximately $0.38 per $1,000 of revenue for top performers and $0.92 for bottom performers. These are cross-sectional performance differences, not evidence that buying a particular technology causes the gap. APQC, reducing AP costs
A company's own baseline remains necessary. Its supplier mix, transaction complexity and cost allocation may differ substantially from the benchmark population.
Prevent repeat exceptions
Once the queue is visible, ask which cases should exist at all. If a department repeatedly orders without a purchase order, a faster investigation may still leave the underlying problem intact. If receipts are entered late, a receiving-process change may remove the discrepancy before AP sees it.
Track recurrence by cause and owner. A useful improvement can reduce both the number of exceptions and the effort needed for the ones that remain. Those two effects should be measured separately to avoid attributing every improvement to AI.
For a pilot, compare a representative period before and after the change, controlling as far as practical for volume and case mix. Include a complete close cycle if that is when the difficult work appears. Record any concurrent policy or staffing changes.
Where Curia can contribute
Curia's assessment starts with the actual work and the client's economics. In AP, that means following cases through the surrounding systems and teams, then identifying where redesign and an operated workflow could reduce effort or delay.
The proof should demonstrate correct resolution, retained evidence and a measured reduction in total human work. A higher automation rate may follow, but it should remain an explanatory measure rather than the sole objective. See Curia vs AP automation software for the broader purchasing comparison.
Bring Curia your AP dashboard and a sample of unresolved cases. We will assess what the automation percentage leaves out and where the next improvement should focus. Review your AP workload