Everyone's Talking About AI in AP. Here's What the Data Actually Shows.

Just like every other department, accounts payable has spent the last few years absorbing a steady stream of AI hype. New platforms, new features, new claims about what's now “AI-powered.” What's harder to find is a clear picture of where AP teams actually stand today: what's broken, what they're doing about it, and where they think AI will actually help.

During a recent webinar on AI-native accounts payable, we polled the live audience, a mix of AP managers, controllers, and finance leaders across manufacturing, retail, healthcare, and services. The sample was small, informal, and specific to that room. But the pattern in the answers lines up with what we hear constantly in the field, and it's worth pulling apart.

Manual exceptions are still the default failure mode.

Asked to name their biggest AP pain point, 53% of respondents pointed to manual exception handling, more than double the next closest answer (slow approval cycles, at 21%). Supplier compliance issues and lack of invoice visibility rounded out the list.

This tracks with a pattern that shows up across most AP automation deployments over the last decade: the straightforward invoices get automated, and the exceptions get left for a person to sort out by hand, every time, indefinitely. Automation solved the easy 80% of the volume and left the hard 20% almost exactly where it was. The interesting part isn't that exceptions happen. It's that the same categories of exceptions tend to repeat, month after month, across the same set of suppliers, without the system ever seeming to learn from the pattern.

Most AP teams have adopted AI features. Very few have adopted AI-native systems.

When asked how their organization would describe its use of AI in AP today, 43% said they're using point features like OCR or auto-coding. Only 24% said they're actively evaluating AI-native platforms, and just 14% said they already have one running.

The distinction matters more than it sounds. “AI-powered” typically means a machine learning feature has been added on top of an existing workflow engine, still built around the same rules-based architecture AP systems have used for 20 years. “AI-native” means the system itself, including the workflow logic, was designed around the assumption that AI would be doing part of the reasoning. Those are architecturally different starting points, and most of the market is still in the first category. That's not a criticism of anyone's roadmap. It's simply where the technology has been until recently.

Touchless invoice rate is the metric under the most pressure.

Half of respondents said touchless invoice rate is the metric they're under the most pressure to improve this year, ahead of days payable outstanding (31%), cost per invoice (13%), and cash visibility (6%).

Touchless rate tends to function as a proxy for the health of the whole AP process. A low touchless rate usually means high exception volume, incomplete supplier data, or workflow logic that can't handle edge cases without human intervention. Teams that are focused on this metric are, in effect, focused on the root cause rather than any single downstream symptom.

Supplier experience is a known gap, not a solved problem.

Asked how central supplier experience is to their AP strategy, no respondents said it was core to their roadmap. The answers split almost evenly between “not really a factor” (30%) and “we think about it but don't measure it” (30%), with the largest group (40%) calling it a growing priority.

This is arguably the most telling result in the set. AP has always been a two-sided process, buyer and supplier, but almost every AP platform built over the last two decades was designed exclusively around the buyer's workflow. Suppliers were treated as an input, not a stakeholder. The result is a blind spot: most finance leaders have limited visibility into where their own suppliers get stuck, which means the same friction gets reintroduced by different suppliers every month, with no mechanism to catch it earlier.

What the pattern suggests.

None of these four results are surprising in isolation. Together, they describe an industry that knows roughly where its problems are, manual exceptions, an unclear AI strategy, a metric leadership is pushing on, and a supplier relationship nobody has fully built for, but hasn't yet had a platform designed around solving all four at once.

That's the gap AI-native architecture is built to close: not a feature added to an existing system, but an AP platform designed from the ground up to prevent exceptions before they recur, extend visibility to both sides of the transaction, and treat the supplier relationship as core infrastructure rather than an afterthought.