PEX CFO Luke Pritchett is seeking to leverage artificial intelligence to build a "shadow ledger" for the company.

The goal is not to replace PEX's existing financial processes, but to create a "parallel version to the accounting ledger," run autonomously by AI agents, operating alongside the company's books as another avenue for Pritchett, as the finance chief, to check for errors. He told CFO Dive in an interview that he hopes to have the shadow ledger 80% to 90% complete by the end of this year.

Founded in 2006, PEX providesspend management solutions, including corporate cards, receipt management, and expense control.

Starting the AI journey

Pritchett's personal AI journey began earlier this year when he used a tool to create an agent to help manage his children's schedules. "That was the first time I really used an AI model, and it changed the way I think about working at PEX," he said.

According to his LinkedIn profile, he first joined PEX as a senior vice president in 2015 and became CFO in 2022. Before joining the New York-based spend management company, he spent seven years at card network Visa, serving in roles including senior account manager on its prepaid card partner development team.

Over the past decade at PEX, he has helped the finance department evolve from its early stages to its current level of serving thousands of businesses. The company is also driving broader adoption of AI.

In July, the company raised$160 million in debt and equity financing, led by Bluff Point Associates. According to a company blog post at the time, the funds were planned for three areas: expanding its charge card program, investing in employees, and "serious investment in AI."

As an AI enthusiast, Pritchett said when thinking about the technology, "It's crazy how much my job has changed in the last six months," compared to the previous nine years, which saw little change. After successfully applying the technology in his personal life, he began using it to address challenges or bottlenecks in PEX's financial processes.

The first AI agent he created was an internal expense policy agent. Finance team members used to spend a lot of time hunting for missing information or receipts in employee expense reports, and the agent "really handled that follow-up step."

"That way, when month-end comes, most of the information has already been collected from employees, and the back-office team can focus on handling the remaining month-end items," Pritchett said. He also mentioned that the company recently completed rolling out a tool related to that agent—a lost receipt affidavit,which can be uploaded when needed, without users having to freeze the associated card—and the tool has been made available to customers.

Beyond saving time for the finance team (which previously had to manually enforce expense policy), employees themselves can also query the agent on Slack to better understand the policy and what they need to provide.

"I think that's probably the most valuable part: not necessarily telling an employee that a transaction doesn't comply with policy, but helping train and teach employees what our policy actually is, in a low-pressure environment where they can do it from their phone or Slack on desktop," Pritchett said.

Building connections

The expense policy agent itself is part of Pritchett's larger project to build a more autonomous, AI-driven back office that uses technology to help share information between different systems. Pritchett has also begun using AI, via Anthropic's Claude supported by the company's enterprise license, to build connections between PEX or corporate card platforms, HR systems, bank data feeds, and accounting systems.

These agents then "move data points between these systems of record, and then I see everything the agents have done on a dashboard, for tracking, to ensure visibility and accuracy, and to make sure it's creating value," Pritchett said.

Pritchett has created usernames and passwords so finance team members, such as PEX's controller, can also access the AI dashboard. As more systems are integrated, the team is gradually becoming more engaged.

As CFO, Pritchett aims to encourage more AI experimentation in a collaborative way. For example, if each employee individually uses generative AI solutions, Pritchett observes "you can get about 20% to 30% more work done." He said: "That's great productivity. Worth having. Fantastic, but it's a bit linear and it hits a ceiling because the work is still trapped where it's being done."

He is currently thinking about how to break through that limitation; if each employee "really has a specific agent, then there could be more inter-agent collaboration," he said. "I think that's when you can break through the next ceiling of productivity and excitement."