What counts as an AP exception?
Anything that breaks the automated path: an invoice that does not match its purchase order on price or quantity, references a missing goods receipt, arrives without required fields, hits a vendor record that does not quite match the vendor master, or gets misread during capture. Each one drops into a queue where someone has to establish what happened.
What is a good AP exception rate?
In the 2026 distribution, the median sits at 11–14% of invoices, the best quartile runs below roughly 8%, and the bottom quartile above roughly 17% (SSON R&A Metrics Intelligence Hub, 2026). The number needs context because widening tolerance bands can lower the exception rate while raising leakage: discrepancies that used to be flagged now pass clean (SSON/Xelix, 2026). A good exception rate comes from cleaner inputs and smarter matching, not looser rules.
The guide to duplicate and near-duplicate payments maps one of the most common paths from an exception to a loss. The review model in the FAQ explains what should happen when a case remains ambiguous.
What do exceptions cost?
The median cost per invoice processed runs from $3 to under $4, and nearly half the market pays more (SSON R&A, 2026). Exception handling is a major reason. Each case can require someone to pull the purchase order, find the delivery note, contact the requester, and chase the vendor. At 11–14% of hundreds of thousands of invoices, teams must choose between processing exceptions and investigating them. Under backlog, processing usually wins.
How do exceptions become leakage?
An invoice in manual handling has lower visibility and a longer cycle time than one in the automated flow, which lets errors compound (SSON/Xelix, 2026). A resubmitted invoice may be re-keyed and re-approved as new, discrepancies may pass without full validation, early-payment discounts can expire in the queue, late-payment penalties can accrue, and a plausible fake has more room to survive.
How do you lower the rate without losing control?
Fix the inputs and upgrade the matching, in that order. Vendor-master cleanup and purchase-order discipline remove exceptions at the source. Then match against the documents instead of widening the bands. An agent that reads the invoice, delivery note, and contract can investigate a price variance using the underlying evidence and assemble a worked case for anything it cannot resolve.
In a cloudsquid customer deployment, agents audited 100% of one year's AP transactions in under two hours. The same checks can run continuously in Accounts Payable, with finance approving worked cases rather than opening raw ones.