A 12-point gap separates the average finance organization in Basware’s network from its best performers when it comes to running the invoice lifecycle without human intervention.

Basware’s inaugural 2026 Finance Performance Report puts its network-wide Lifecycle Autonomy Rate at 81%. Among the top 5% of qualifying customers for that measure, the rate reaches 93%. The finding comes from production data rather than a survey: data from more than 2.5 billion invoices processed across 6,500 enterprise customers in over 190 countries, representing more than $10 trillion in combined spend.

That scale gives the report a useful way to look at a question finance teams have spent years wrestling with. Automation is already common, so why do some organizations still require substantially more human intervention than others?

Part of the answer is that automating individual tasks and allowing an entire process to run with limited intervention are different accomplishments. Invoice capture can be automated. Matching can be automated. Coding can be automated too. The harder job is getting those pieces to work together reliably enough that people spend their time on the cases that genuinely need them.

Automation Can Plateau Long Before the Work Disappears

Accounts payable has been automating routine work for years, and clean invoice flows are no longer the hardest part of the problem. The stubborn work sits in exceptions: mismatched data, unclear coding, unusual supplier behavior and the cases where a simple rule cannot confidently determine what should happen next.

Basware’s report measures autonomy across the full invoice lifecycle, from receipt and matching through transfer. Its 81% network average suggests most of that path can already proceed without manual handling. The 93% best-in-class figure shows there is still considerable room between having automated steps and having an operating model where those steps hold together.

That difference also appears in the report’s AI Decision Rate. Across the network, 89.4% of AI-driven coding and approval-routing decisions are completed correctly without human intervention. Among best-in-class organizations, the figure is 96.5%.

A seven-point difference in decision reliability can have an outsized operational effect when the underlying process handles hundreds of thousands or millions of invoices. Every decision the system cannot make confidently has to go somewhere, usually back to a person.

Donna Wilczek, Basware’s Chief Product and Technology Officer, argues that higher autonomy does not require giving up control.

“Best-in-class organizations aren’t choosing between automation and control,” Wilczek said when the report was released. “As AI transforms finance, that kind of visibility is critical to knowing where to automate, where to increase trust around AI, and where to keep people in control.”

The Better Systems Appear to Need Fewer Rescue Missions

The report gets more interesting when autonomy is compared with the amount of work being pulled back for review.

Across Basware’s network, 1.4% of invoices are flagged as potential fraud or duplicates and sent for manual review before posting. Among the best-performing organizations on that measure, the figure falls to 0.05%. Basware interprets lower rates as evidence of stronger controls earlier in the process, meaning fewer questionable invoices make it far enough to require intervention.

That points to a less obvious way of measuring AI performance. A finance team should not necessarily celebrate because its system is escalating thousands of suspicious cases. If weak supplier data or poor controls are creating unnecessary exceptions, the organization has simply moved manual work further downstream.

The strongest operations appear to prevent more of those cases before they become exceptions.

Payment performance shows a similar gap. Basware reports that best-in-class organizations pay 92% of invoices on or before the due date, compared with 81.8% across the broader network. It also found that the leading organizations receive 99.7% of invoice volume through a single centralized compliance platform.

Those numbers move the conversation beyond how many invoices a system can process without somebody touching them. Finance performance also depends on what happens because of that automation: whether suppliers are paid when expected and whether compliance becomes easier to manage across countries.

Autonomy Is Becoming a Reliability Problem

The next phase of finance automation may be less about finding another manual task for AI to perform and more about raising confidence in the decisions already being automated.

“Confidence in those decisions comes from what the AI has learned from,” Wilczek said. “Ours is network-trained AI, or Networked AI as we call it at Basware. It has learned from invoices across 6,500 companies, not one. An AI trained only on your own invoices has never seen the supplier, the format, or the fraud pattern that is new to you but routine somewhere else in the network.”

That changes the human role too. People spend less time executing routine transactions and more time deciding where AI has earned the authority to act without them. When performance slips, they need enough visibility to understand why.

For now, the data suggests that finance teams have moved beyond proving automation works. The harder benchmark is how rarely people have to step back in.