Most invoices aren't waiting on money. They're waiting on a reply.
In most mid-market finance teams, the bottleneck in accounts payable isn't the payment run. It's the chasing. Where AI genuinely helps, and where a person still has to sign.
Ask a finance lead where supplier invoices actually get stuck, and almost none of them will point at the payment run.
The money is the easy part. An invoice arrives, usually by email and occasionally still on paper. Before anyone can pay it, it has to be matched against a purchase order, coded to the right cost centre, and approved by the person who ordered the thing. That last step is where it sits. Often for days.
The approver is on site, or on leave, or doesn't recognise the line item and wants to check before they sign off. So the invoice waits. The supplier rings. Someone in finance forwards the email again with "any update on this one?" and goes back to the other forty.
None of that is a data-entry problem. It is a coordination problem. And coordination is the kind of work that quietly eats an afternoon without ever looking like a task you could point at.
Where the time actually goes
If you watch a busy accounts payable inbox for a week, the manual effort clusters in a few predictable places:
• Reading each invoice and pulling out the supplier, amount, PO number and dates.
• Matching it to the right purchase order and the goods-received note, then spotting the ones that don't line up.
• Working out who has to approve it, and chasing them until they do.
• Answering the supplier who wants to know when they will be paid.
Three of those four are reading, matching and chasing. They are repetitive, they follow rules, and they are a large part of why the close runs late.
What AI is genuinely good at here
This is the part of finance where the current tools earn their place. A system can read the invoice, extract the fields, and match it to the purchase order and the delivery without anyone typing. When the three line up, it can prepare the invoice for approval and route it to the right person automatically. When they don't, it can flag the exception with the reason attached: price variance, missing PO, quantity mismatch.
It can also do the chasing, which is the part people hate most. A polite, accurate reminder to the approver. A status reply to the supplier. The thread kept alive without a person having to remember it exists.
What is left for the finance team is the work that actually needs judgement. The genuine exceptions. The new suppliers. The invoice that looks wrong because it is wrong. That is a better use of a qualified person than re-forwarding emails.
Where a person has to stay
Paying money out of the business is not a place to remove the human and walk away.
Keep approval with a person, especially above a sensible threshold. Let the system prepare, match and recommend, but have someone sign off the payment. Log what the system did on every invoice, so that when an auditor or a supplier asks, you can show the trail. And actually read the exceptions it raises rather than waving them through, because a tool that flags a duplicate invoice only helps if someone looks.
The aim is not an accounts payable function that runs without people. It is one where the people are spending their time on the invoices that need a brain, and not on the ones that just needed a reply.
So if your month-end runs late and the assumed fix is more hands, look first at where the invoices are actually waiting. In most teams I see, it isn't the money. It's the reply.