Every swipe, ACH transfer or virtual card transaction carries clues about who is spending, where money is moving, what is being bought and whether something looks wrong.
Those clues have remained relatively untapped, and not for lack of enterprise effort.
“The core problem is work happens in one place, and payments happen in another. People are stitching that together manually, which can create delays and errors as volume grows,” Ryan Taylor, SVP Product Management – Mobility & Payments at WEX, told PYMNTS during a discussion for the PYMNTS B2B Payments Event 2026 — Stopping B2B Fraud Before AI Moves the Money.
Artificial intelligence, Taylor said, could eliminate some of that stitching. But the bigger opportunity is to collapse the distance between detecting what should happen and safely executing it.
After all, AI systems are getting better at determining which invoice looks suspicious, where a fleet vehicle should refuel, whether a supplier price is drifting higher and which transaction deserves intervention. But identifying the right action is different from having the authority to take it, particularly when taking it means moving real corporate cash.
And that could just end up turning payments from the endpoint of an enterprise workflow into its control plane.
“Payment data is interesting because it really is the digital footprint of business,” Taylor said. “It tells you more than just what was spent. It can tell you about what was happening in the business at that exact moment that a payment occurred.”
AI Can Make the Decision; Payments Will Decide Whether It Can Act
The next bottleneck for enterprise AI is shaping up around permission, not intelligence. And payment infrastructure can provide something models cannot generate themselves: enforceable boundaries around what an AI system is actually allowed to do.
Knowing that a driver spent $100 on fuel is accounting data. Combining the card swipe with the vehicle’s location, fuel level, route and historical behavior creates operational context. An AI system can potentially determine whether the purchase is legitimate, whether the amount makes sense and whether the driver could have fueled more cheaply elsewhere.
Taylor argued that effective artificial intelligence systems therefore require three elements: context, controls and clarity. He compared the process with onboarding an employee.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
“You need to give them context. You need to put them in the right control environment so they can learn and grow, but not do harm. And then they also need clarity. They need to know what success looks like,” Taylor said.
The enterprise AI race, as a result, could eventually become partly an identity-and-permissions race: Which agent is acting? On whose behalf? What authority has it been delegated? Which counterparties can it transact with? How much can it spend? When does it need human approval?
The Virtual Card Takes on a New Role in an Agentic Landscape
Virtual cards can already restrict merchant, amount, timing and number of transactions. In an agentic environment, those features effectively become programmable permissions for machines.
“One of the main reasons virtual cards are so compelling as we look towards an agentic future is that controls are built directly into the payments themselves,” Taylor said. “We don’t want to give the agents open-ended access to our corporate bank accounts. We need to govern how and when they can execute payments, and how money is actually moving to manage the risk exposure.”
That ultimately suggests the future enterprise payments stack could function less like plumbing and more like an API for financial authority. An agent might determine that an invoice should be paid. Payment infrastructure determines whether it can pay it.
That division of labor could also change how companies measure AI’s financial value. The obvious efficiency case is fewer manual tasks across areas like invoice matching and reconciliation, where finance teams can automate work and redirect employees toward higher-value activities.
The deeper opportunity is moving financial controls upstream. Much of corporate finance has operated retrospectively, but AI paired with real-time payment data can move controls around exceptions, fraud, and variance closer to the moment when money changes hands.
Taylor describes the shift as “shrinking the distance between this insight to action.”
“The organizations that get the most out of AI won’t be the ones that add chatbots to legacy systems,” he said. “They’re the ones who are going to treat payments as a core infrastructure.”
Watch the full PYMNTS B2B 2026 Event interview with WEX’s Ryan Taylor to hear more about:
- Why AI’s next payments bottleneck isn’t intelligence. It’s permission.Taylor says AI can increasingly identify what should happen next, but enterprises still need hard limits around what machines can actually do with corporate money — turning payments into the control layer between AI decisions and financial action.
- Why the virtual card could become the wallet for enterprise AI.Taylor says virtual cards can restrict who gets paid, when, where and how much, giving AI agents the ability to execute transactions without giving them “open-ended access to our corporate bank accounts.”
- Why AI could move finance from catching mistakes to stopping them.Taylor says combining payments with operational data can shrink the distance between “insight to action,” letting businesses detect fraud, invoice discrepancies and inefficient spending before money moves instead of finding problems during reconciliation.
Add as Preferred Source
