Where AP automation projects stall
AP automation projects stall at the exception queue. Capture, e-invoicing, and workflow tools push clean invoices to touchless processing within a few quarters, and then progress stops, because the invoices that remain fail matching for reasons that live outside the AP system: wrong PO lines, missing receipts, prices that changed mid-contract, supplier records that disagree across platforms. The project plan assumed that work would shrink along with everything else, and it doesn't.
The pilot automates work that was already cheap
A large majority of invoices arrive with a valid PO, match on the first pass, and post without anyone touching them. Those invoices were always the cheap ones. Capture and workflow tools make them cheaper still, which is why the first six months of a rollout produce such a satisfying chart. The business case, though, was built on the smaller and far costlier share: invoices that fail the match. Those still route to a human queue, just as they did before the project. If you are evaluating tools, our practical guide to AP automation covers what each layer of the market does; the short version is that most of it is built to make cheap invoices cheaper.
The exception queue outlives the go-live
Exceptions are created upstream of AP, which is why an AP tool cannot close them. A price that changed after the PO was cut, a receipt that was never posted, a supplier invoicing from a different entity than the one in the vendor master: each of these needs information from procurement, receiving, or the supplier before anyone can approve payment. (The full catalog is in why invoice exceptions happen.) Exception volume also varies enormously with how well that upstream process is run: Ardent Partners' research puts Best-in-Class AP teams' invoice exception rates roughly 59% lower than the rest of the marketplace. A team on the wrong side of that spread carries its full queue through go-live, keeps the analysts who work it, and discovers that the exception queue is a headcount plan with better software attached.
Per-seat pricing rewards a full queue
Some of the stall is structural. A vendor that licenses its software per seat is paid for every person who works the queue, so a smaller queue means a smaller renewal. That vendor is glad to help you process exceptions faster; helping you have fewer of them would shrink its own contract. The same logic now applies to "agentic" products sold per seat, a contradiction worth sitting with – the vendor of an agent priced like a person needs the queue to stay staffed. We wrote about this incentive in per-seat pricing and the incentive to keep queues full.
Phase two stalls on integration debt
Phase one connected one system, usually the ERP. Phase two, the part that would touch exceptions, needs receiving data from the plant system, contract terms from a repository, supplier records from a portal, and half its context from email threads. Every connector is its own IT project with its own security review, and rule-based tools demand standardized data before they can act on any of it. So the roadmap slips a quarter at a time until the sponsor moves on, and the project dies in a backlog without anyone ever deciding to kill it.
A stalled project restarts at the exception queue
The way back is to aim the next dollar at the exception work itself. That takes software that can read across ERP, portals, spreadsheets, and email the way an analyst does, resolve routine cases end to end, and hand judgment calls to a human with the evidence already assembled – nothing ripped out or replaced, and no data standardization project standing in front of it. When that works, the payoff is capacity: the analysts who were keying status updates into a tracker get their hours back for supplier disputes, early-payment discounts, and the root-cause fixes that shrink the queue at its source.
It helps to notice who benefits from the stall: the BPO working your queue bills per head, the consultants who scoped phase two bill by the hour, and the per-seat vendor renews a license for every analyst still in the queue. Fragment is built to come out ahead only when the queue shrinks – its agents resolve exception work across the full source-to-pay lifecycle, from PO creation to tax holds, on top of the systems you already run and in production within weeks. The workflow catalog shows which queues they already handle, and a demo is the quickest way to watch one work a real exception end to end.