How CFOs Can Handle Employees’ Expensive Taste in AI

Everyone’s ordering AI, and they’re all putting it on the CFO’s tab. 

Now, those running finance teams and leading communities of finance professionals say the strongest CFOs are setting firm controls on AI spending and making teams prove it’s paying off.

“One heavy user can consume more tokens than a hundred colleagues,” Omar Choucair, CFO of financial close software provider Trintech, told CFO Upside. “Without limits, a full year of budget can be utilized in a month.”

For Chris Argent, founder of finance leadership community GENCFO, the spending question is part of a much broader challenge. He said finance leaders recognize the opportunity to redesign work, but implementing that change is proving harder than agreeing it should happen.

“Many leaders understand this shift but are much less clear on implementation,” Argent told CFO Upside.

Whose Budget Is It Anyway?

Companies have experience centralizing analytics and master-data management, Argent said. But AI adoption is spreading less neatly, making it harder to keep track of spending across the business. “Right now AI is everywhere, which means ‘shadow IT’ costs are everywhere,” he said.

Assigning ownership by department appears to be a starting point, according to Argent, though that structure could change as companies better understand the technology’s costs and capabilities. Proving a return remains difficult when the use cases themselves are still developing.

“AI use cases are still emerging, so we can not lean on ‘tried and tested’ finance transformation outcomes,” he said.

Choucair said CFOs often inherit responsibility for AI spending without a clear agreement on who owns the results. “In many companies the AI budget lands with the CFO by default rather than by design,” he said. “If nobody clearly owns it, nobody really owns the outcome.”

Trintech established an AI task force earlier this year with representatives from every department, with tasks to review usage data weekly and reports to executives every two weeks, Choucair said. The company also puts limits on token spending.

His funding threshold starts with familiar questions for CFOs: What problem does the investment solve, what does it cost and what savings can the company reasonably expect? “If we can’t describe the problem clearly and are unable to diagram the data feeds and estimate the anticipated savings, it typically will not get funded,” he said.

Pending Human Review 

Getting a project approved does little to guarantee employees will change how they work. Argent described a global hotel chain with a loyal workforce and high employee net promoter scores whose teams were nevertheless protecting their positions and resisting change. In another company, he said, a regional CFO’s efforts to address process problems ran into resistance from the board itself.

“Even the board sees the CFO’s progressive ambitions as outside the CFO remit, when it is clearly the remit of all modern leaders,” Argent said. A CFO expected to justify an AI investment may also need explicit backing to change the processes determining its return.

At Trintech, rolling out enterprise AI licenses revealed different levels of employee readiness, Choucair noted. “Our people sorted themselves into three groups: eager champions, people worried they lacked the skills and a large middle that needed training,” he said. “That is why we treat building the capability internally as seriously as hiring for it.”

The return calculation also has to account for what happens after AI completes its assigned tasks. Trintech’s accounting team, without surprise, uses its own AI platform for reconciliations and close management. Choucair shared how agents gather evidence and recommend next steps, while employees review the output and then make final decisions.

“The best way to improve AI’s ROI is to design verification into the process,” he said. “When the evidence and audit trail arrive with the output, the financial and operational review moves faster, and the team spends its time on the exceptions instead of rechecking everything.”

Even a confidence score needs scrutiny, as Choucair recommends tracking how often employees overturn findings the system rated highly. “The score should shape how much review an item gets, not whether it gets reviewed,” he said.

Choucair also cautioned against tying compensation directly to AI usage, saying employees should be evaluated on what they accomplish with the time automation frees up. Argent expects verification and exception management to become core finance skills, increasing the value of employees willing to keep learning.

Check, Please: Choucair’s test for whether teams are ready to rely on AI is straightforward: Can they explain its output to their auditors? “If they are unable to, the AI tool isn’t ready for full integration, regardless of what the score says.”