Sit through enough AP automation demos and you'll notice they all open the same way: a slide about extraction accuracy. "We read invoices at 99% accuracy." A confident drag-and-drop of a clean PDF, fields populating in real time, a satisfying green checkmark. It's a great demo. It's also a tell.
Vendors lead with extraction because it's the part that demos beautifully and the part that's easiest to claim a number on. But it's worth asking why the demo starts there — and whether it starts there because extraction is where your AP team actually spends its day. For almost every team, it isn't.
Extraction is the part that was already mostly solved
Reading data off an invoice is real work, and modern AI does it well. But it's the opening move of accounts payable, not the game. Once the data is captured, the operation that consumes your team begins:
- Coding non-PO invoices to the right account, cost centre, project, or fund — where there's no PO to lean on and judgment is required.
- Working the exception queue — the invoices that fail matching, the price mismatches, the missing receipts, the things that need a human to chase a fact down and resolve it.
- Routing for approval the way your organization actually delegates authority, with each approver seeing only what's theirs.
- Handling the supplier inbox — status chasing, missing PO numbers, duplicates, banking-change requests — the conversation that happens before an invoice is even clean enough to enter a workflow.
None of that shows up on the accuracy slide. All of it is where the hours go. We've made this case from the technical side in why Oracle Fusion teams still add a dedicated AP layer even though IDR already captures invoices — capture is roughly the first 20% of AP, and the demo spends 80% of its time there.
Why the accuracy number is a weak basis for a decision
Even on its own terms, the headline number is hard to compare. "99% accuracy" measured on whose invoices? Clean PDFs from large suppliers, or the smudged scans and oddly-laid-out one-off vendors that make up the messy tail of your supplier base? Field-level or document-level? On the formats you actually receive, or the demo set? The number is real but rarely apples-to-apples, and chasing the last point of extraction accuracy delivers almost nothing if the time sink was never extraction in the first place.
More importantly: extraction errors are visible and self-correcting — a wrong field gets fixed in review and, in a good system, fed back to improve future accuracy. Exception-handling gaps are invisible in a demo and chronic in production. You don't find out the exception queue is unworkable until you're three weeks into go-live.
What to actually ask in the demo
Shift the demo away from the happy path. The questions that separate real AP automation from a capture engine with a workflow attached:
- Show me the exception queue. When an invoice fails matching, what does the person who has to clear it see? Do they get the PO, the receipt status, the supplier context — or just a flag and a blank screen?
- What happens to a not-receipted PO invoice? Does it sit in AP, or does the system know to route the receiving action to the requester who can actually clear it?
- Show me the supplier inbox. When a supplier emails "did you get my invoice?" what happens? Is it a person typing a reply after checking the ERP, or is that answered automatically from live status?
- What about the non-invoices? Statements, sales orders, duplicates, junk — how do those get sorted before they enter the workflow?
- How does it post into our ERP? For an Oracle shop specifically: Oracle-supported methods (FBDI, OIC REST), or custom base-table writes you'll have to defend at every upgrade?
The answers to those questions predict your real workload far better than any accuracy figure.
The inbox is the part nobody demos
The single most telling question is about the supplier inbox, because it's the work that's least visible and most relentless. The emails that arrive before a clean invoice exists — the missing PO, the wrong remit-to, the "any update?", the banking-change request — never appear in an extraction demo, yet they're a steady drain on the AP team every single day. A tool that resolves those upstream changes the math entirely; a tool that ignores them leaves your best lever untouched. That's the whole premise of supplier-email exception prevention, and it's exactly the part a capture-led demo will never bring up.
Where EZ Cloud fits
EZ Cloud does extraction well — but we'd rather you evaluate us on the parts that actually consume your team. The exception queue with live ERP context. The not-receipted hold routed to the requester instead of AP. The supplier inbox handled by AI email exception prevention before anything becomes an exception. And clean posting into Oracle EBS and Fusion using Oracle-supported methods, so your upgrades stay boring.
When you're evaluating AP automation, let the vendor have their accuracy slide — then ask to see the exception queue and the inbox. That's where you'll find out whether you're buying automation or a better demo. When you're ready to compare on what matters, see how the pieces fit.