Finance teams could use a refund on some of that promised free time.

Years of technology spending have left finance teams with plenty of manual work, and AI is arriving with another promise to clear their desks for the highly coveted operational duties. Whether it delivers, finance leaders say, depends partly on how much work remains after the agent finishes.

The monthly close is one process where promised relief has been slow to arrive, according to Alyona Mysko, founder and CEO of Fuelfinance and a former CFO. “[Many] CFOs paid for an ERP to solve it, and the manual work is still there,” she told CFO Upside. “Having the system in place doesn’t mean people stopped doing the controlling by hand.”

At Prophix, a provider of financial planning, budgeting and reporting software, CFO Aaron Levine sees fragmentation persisting after companies moved finance onto cloud platforms. “You can’t throw an LLM on top of disconnected data and expect the kind of consistent, reliable answers finance requires,” he told CFO Upside.

Data, Duh

Mysko said AI rollouts can stall when data needs attention and responsibility for that work is unclear. “The data isn’t ready, and nobody owns fixing it,” she said.

When systems produce conflicting numbers, she also added that teams need to decide which one to use. “When you have two revenue numbers in two systems and no rule for which one wins, AI gets confused exactly where a person would,” she said. “You have to write down the rules and decide where the final number lives.”

Before an agent can take over a task, she said the process also needs to be documented. “Is the process written anywhere, or is it in someone’s head?” Mysko asked. Implementing agents often means standardizing processes first, she said, sometimes changing them entirely if required.

Within his own finance team, Levine said AI’s payoff comes from reducing the time spent preparing information for analysis. “The biggest gain has been in the manual, repeatable work around the forecast, not the forecast logic itself,” he said.

To help explain variances and update forecasts, the team uses agents that also gather input from budget owners. That means less time spent chasing people for updates, while calculations stay within the model. “Every action still goes through human approval, which keeps the process auditable,” he said.

Buying Time

If checking AI’s work takes as long as doing it, the promised time savings become hard to find. “There’s no return if you switch from doing every line to checking every line,” Mysko said. “That’s the trap.”

To reduce that burden, Mysko recommends audit logs and clear rules about which exceptions need human attention. “If you used to check 3,000 rows doing the same task on repeat, you now check 10,” she said.

“A process that used to eat a chunk of someone’s day now gets reviews in minutes,” Levine said of automated bank reconciliations. His team still reviews the results, but, like in his forecasting, some of the work leading up to the review takes way less time.

When explaining why human review still matters, Levine compares AI to a familiar hire on the finance team. As he put it, “think of AI as a junior staff accountant. It still needs review, but we need junior staff accountants to handle some of the mundane work to free us up to focus on more strategic areas.”

Even after an agent starts handling tasks, Mysko pointed out, someone has to keep it running. “When you build agents yourself, in Claude or similar, you’re building software,” she said. “Someone on the finance team has to support it, update the sessions and maintain the context.”

Earning Its Keep. For Levine, the return on an AI investment depends on how much work it takes off the team’s plate, including the upkeep it requires. “My first question is always whether it will actually give finance teams more capacity,” he said. “Any AI tool worth its salt needs to prove that it’s not just another system that finance teams have to actively maintain.”

Has an AI-assisted answer delivered by a cheeky analyst come through your door lately?

At recent CFO Alliance roundtables, finance leaders have discussed how AI is giving employees answers to workplace challenges, sometimes with enough confidence to bring those answers straight to the CFO’s office.

Whether that confidence survives a few follow-up questions is where things are now getting interesting.

Nick Araco Jr., founder and CEO of the CFO Alliance, sees that gap becoming a concern in conversations with midmarket finance leaders. Following the organization’s Q3 roundtables in cities including Denver, Philadelphia and New York, as well as sessions in Southern California and Nashville, he said the discussions were previously about how teams should use AI. Now, it’s about how CFOs should manage adoption already happening without their approval.

“Six months ago it was ‘should we?’ Now it’s ‘it’s already happening and I didn’t sign off on it,’” Araco told CFO Upside. Members are also questioning whether employees can evaluate the work produced by shadow AI and whether their managers are asking enough questions.

CFOs are taking steps to address unauthorized use, Araco explained. He said in one session, members developed a working AI data standard together. “CFOs aren’t waiting for permission [to establish working standards for AI use] anymore,” he said.

Cloaking a Skills Gap

Araco calls the problem the “AI Confidence Curve,” his framework for describing how users can overestimate their abilities when AI produces convincing work. The risk, he said, becomes apparent when a finance employee can deliver an analysis without understanding how to build it.

“The output is fluent, but fluent isn’t the same as right,” Araco said. “And the analyst who can’t build the variance analysis from scratch can’t tell when the AI got it wrong either.”

Araco’s CFO members have suggested to each other that with AI doing more of the work, it’s now up to the executive leadership team to take a closer look at how employees are learning. Nowadays, a finished assignment may tell a manager less about what someone understands, making a few follow-up questions very useful.

“Confidence goes up, competence quietly goes down and nobody notices,” Araco said.

All Talk

Araco said he sees a similar communication problem in how finance chiefs assess their leadership. A CFO may believe a decision was explained clearly while employees still have questions. So his next round of sessions in Q4, which includes 21 cities in 65 days through mid-December, will ask participants to test that assumption with their own teams.

“Most of us rate ourselves as strong communicators. Few of us could prove it,” he said.

Before attending a Q4 roundtable, each finance leader will ask their team: “What’s one thing I think I’ve communicated clearly that you’re still unsure about?” Participants will bring the answers to the session, where other finance leaders will help them compare that feedback with how they rate their own communication.

Then, each participant will choose one gap to address in 2027. A fellow participant will follow up to check on their progress, giving the commitment real accountability after the session ends.

Where the Buck Stops. “And when your team stops questioning the tool and stops questioning you, errors move faster, and nobody says a word until the CEO, board or investors do,” Araco said.

Last year, Steve Song traded lobster rolls for flame-broiled burgers.

As the former CFO of Luke’s Lobster, a seafood restaurant company with a global wholesale business, Song managed a working capital structure that required paying fishermen in cash within three days while waiting weeks to collect from customers.

Now as CFO of Rackson Restaurants, Song oversees finance for a franchise operator with nearly 100 Burger King locations, a few Dave’s Hot Chicken restaurants and more than 1,500 employees across the Northeast, mid-Atlantic and Cleveland, Ohio, area. Now, keeping cash available for new restaurants and remodels is a central part of the job.

“The difference between Luke’s and Rackson is I don’t have the wild swings of seasonality,” Song said. “I don’t have the working capital dynamic I did at Luke’s, where I’m paying out vendors 30-60 days ahead of collecting on AR.”

Whether his product comes from land or sea, liquidity remains a priority. “I’ve always said cash is the oxygen of a business,” Song said. “If you don’t have cash, you can’t keep going.”

A Fresh Coat of Capital

Building restaurants and remodeling older locations requires Rackson to set aside cash alongside the recurring expenses that fast food brings. Some Burger King restaurants have older facades that need updating, leaving Song to determine which locations receive funding first. “I need to make sure we have cash flow earmarked for those investment opportunities,” he said.

Expected financial returns help establish one ranking in determining which stores get remodeled, he explained. “If I invest X dollars to remodel, what is the return on that investment per store? What is the lift in sales and profitability?” Song said. “I would probably rank them that way.”

A store’s age can move it up the list, he said, while cash conservation may favor a less expensive project. Under those circumstances, “I might focus on which remodels require the least amount of dollars and go in that order,” Song explained.

By keeping those rankings separate, he gives Rackson’s stakeholders a clearer view of why their priorities differ before identifying investments that satisfy several needs. “Otherwise, the discussion is too commingled with various agendas and interests,” he said.

Song broke down why the franchise relationship also divides some of the responsibility for restaurant performance. Rackson handles site selection and operating expenses, including labor, while the brand oversees marketing and menu development. “There’s a shared responsibility and goal of serving guests, taking care of your team, generating sales and hopefully profit,” Song said. “We just have different areas of focus.”

Taking Orders From Operators

Before introducing changes, Song said it was important to learn Rackson’s systems and understand his predecessors’ decisions. “I think the reputation for new executives is that they come in and want to run their old playbook with the new team,” he said. “I wanted to make sure that wasn’t perceived as the reason changes were being made.”

Restaurant operators and those in direct contact with customers have since gained more influence over back-office priorities. Their feedback identified a need for better labor scheduling, Song said, and more accurate reporting on actual versus theoretical food costs per menu item.

“We want our team’s restaurant operators to be our clients,” he said. “We need to pay attention to our operating partners and truly understand what their teams need, want and expect to do their jobs.”

That feedback gives the technology and other executive decisions a clearer purpose and vision for all of the moving parts the franchise model entails. “Otherwise, we had departments coming up with ‘solutions looking for problems’ and asking operations to follow them,” he said. “That was backwards in my opinion.”

With AI, Song initially urged patience as enthusiasm spread among executives at Luke’s. “It was too early, and it couldn’t do the things they thought it could do, but they were very focused on using AI somehow,” he said.

Now, Song said Claude Cowork changed his assessment process as a CFO because it could perform tasks, an appealing capability for a finance leader with no coding experience. “I didn’t want instructional guidance on how to do something,” he said. “I wanted AI to do it for me.”

Starting with simple Excel tasks, his team progressed toward creating dashboards and building agents. Song said employees are “all increasing their productivity by at least three times, if not more.”

He broke down how Cowork’s ability to execute tasks marked the turning point. “Cowork took a chatbot and gave it the ability to take action,” he said. “When that came out, that was when I really engaged.”

Open 24 Hours. His next ambition extends that work beyond tasks employees actively supervise. “Now our next step at Rackson is figuring out how to create agents that perform work while we’re doing other work or sleeping at night,” Song said.