Your invoice process is mostly exceptions

Ask an accounts payable lead to describe their process and you will hear a clean flow: invoice arrives, matched against purchase order and goods receipt, posted, paid. Then pull ninety days of event data and count how many invoices actually took that path.

In the engagements we have measured, the answer sits between forty and seventy per cent. The rest went somewhere else.

Where they go

No purchase order. Someone ordered by email. The invoice arrives with nothing to match against and goes to a person to chase an approver.

Quantity mismatch. Delivered nine, invoiced ten. Small enough to tolerate, common enough to cost a full-time role across a year.

Wrong entity or cost centre. Correctly issued, wrongly addressed, and now an email thread.

Late goods receipt. Nothing is wrong except sequence - the invoice simply arrived before the receipt was booked. These clear themselves and still consume handling time, because somebody has to look.

Why this matters before automation

An IDP tool sold on straight-through processing quotes its accuracy against the clean path. If the clean path is half your volume, the headline number describes half your problem.

The larger opportunity is usually not extraction at all. It is the late goods receipt that could be waited on automatically, or the tolerance rule that could absorb the nine-versus-ten mismatch without a human. Both are policy changes with an automation attached, and both are invisible unless the exceptions have been counted.

How to count them

Export the invoice document flow with timestamps and status changes. Group by the reason a case left the happy path. You do not need a mining platform for the first pass; a spreadsheet and an afternoon gets you the ranking, and the ranking is what changes the plan.

What usually happens next

Roughly a third of the exception volume turns out to be one supplier, one cost centre or one policy. Fixing that is cheaper and faster than any automation, and it makes the automation that follows substantially simpler to build.