AI growing pains are shattering CFOs’ illusions

Sponsored by

Ready to move past the Wild West of AI experimentation? Try this decision-making framework that PwC’s experts came up with. It outlines where to lead, lag, or exit on AI strategy. Get the details.

When it comes to AI in finance, the hype is hitting the fan.

At the 2026 Gartner Finance Symposium/Xpo, VP analyst Clement Christensen described executives’ prevailing sentiment toward AI as “hopeful disappointment.” During a keynote session, he said “It sounds like this: ‘Yeah, we’ve seen some cool stuff, but it hasn’t translated into any ROI yet.’”

“We’re in the trough of disillusionment,” his colleague and co-presenter, VP analyst Tamara Shipley, said, pointing to Gartner research that shows only one in three AI initiatives increase productivity, while just one in five leads to measurable ROI.

Still, AI’s capabilities as well as companies’ willingness to embrace the technology have increased dramatically over the past year. “The maturity of the models and the tools that are available have exponentially increased,” Rajiv Ramachandran, SVP and CPO of invoice to pay at Coupa, told CFO Brew. He’s getting more granular questions from customers, he said, that suggest “they’re thinking about putting something into production, not thinking about ‘Hey, can I play around with this for another couple of weeks?’”

But as companies move past the pilot stage, problems crop up. CFOs, consultants, and vendors who spoke with CFO Brew at Gartner described many AI growing pains.

Stuck in pilot mode. Many companies are struggling to scale AI past the pilot phase, sources said. “What we are seeing is a lot of success in building demos and pilots, and really low success in going to production and serving a real-life use case,” Yair Weinberger, co-CEO and cofounder of AI startup Reindeer, told CFO Brew.

Companies are also learning that agents aren’t as turnkey as they’re made out to be, Weinberger said. “Building an agent is almost a commodity today,” he noted. “It is actually really easy to build an agent.” The hard part, he said, is maintaining the agent, likening the enthusiasm around agents to a kid who’s excited to bring home a new dog, but not so thrilled to learn they have to feed and walk it.

AI can bring competitive advantages, “but most organizations are not there yet,” Deirdre Ryan, global finance transformation leader at EY, told us. “They’re still dipping their toe in, or they’re doing proof of concepts, or they’re trying to upskill their finance team.”

Data woes. Not having clean and organized data is one key reason AI projects stall. “If you talk to CFOs, you talk to CIOs, you talk to the people in any sector of any size organization, data wrangling is their number one issue,” Sam Ganga, partner and US AI leader at KPMG, said.

“The biggest mistake that I see suppliers making is that they jump right to the AI, and they don’t focus on creating unified data,” Ryan agreed. Clients might have data stored in multiple ERPs and either systems, as well as external data they want to incorporate. “They have to spend the time creating the data model that’s consistently defined, timely, and captures the data elements that you need to drive value,” she said.

Siloed data can also be a problem. At one global pharmaceutical company that Ryan visited, different teams had developed agents that saved manual effort and improved accuracy. The problem was that each team had used its own data to do so. “What the sales forecasting team had done—terrific. What the controllership team had done—fantastic,” she said. “But collectively, they fall short,” she said, because “somebody is going to need to reconcile all those data sources.”

Escalating costs. As AI companies move to token-based pricing, costs are rising. Uber has had to throttle its employees’ monthly AI usage, and even Microsoft canceled many of its Claude Code licenses due to cost.

“CFOs are struggling with how much this is going to cost, because the tokenization is early-stage,” Omar Choucair, CFO of Trintech, told CFO Brew. “It’s a double-edged sword, because you really want your employees to maximize the utilization of these LLM tools, but the dirty little secret is that every week or every month you’re going to get a bill, and it’s going to be significantly more than you had expected.”

That’s “created a completely new analysis for CFOs,” he said, who now are under pressure to determine who in their companies is using AI and how, and what the ROI is.

Optimism remains high. That said, many companies are enjoying productivity gains from AI. “We have customers who are seeing several hours of value added to their business processes, taken away from humans, managed by agents,” Coupa’s Ramachandran said. One company was able to use AI to reconcile five years’ worth of unmatched transactions in around three weeks, Tom Hood, EVP for business engagement and growth at the AICPA, told CFO Brew.

The accounting world is more than ready to embrace automation of rote tasks, according to Hood, a goal “which we’ve been talking about in finance for 20 years.” The AICPA and CIMA’s Rise2040 project, which surveyed around 6,000 accountants worldwide, revealed widespread optimism toward AI and its potential for eliminating what Hood calls “soul-crushing” routine work.

SaaS vendors, naturally, are still picturing end-to-end automation. “What we believe is that the entire process of B2B commerce, of trade, is going to be autonomous, driven by AI agents,” Ramachandran said.

Scott McDermott, CFO of Esker, envisions an almost sci-fi future. “My ultimate goal is not to actually work on my laptop,” he told us. “I should pick up my phone and be able to know directly from an agent how many bookings we had yesterday” or “how much cash we received, which customers missed their payments, how we’re doing versus the forecast, what payments are scheduled.”

Gartner’s analysts feel that AI is ready to take on that higher-value work.

“Using AI to drive efficiency in transactional processes is not the best place to look for ROI,” Christensen claimed during his keynote, as “these processes typically already employ lower-cost labor.” Companies need to “break this mentality” that AI will automate rote work and enable employees to do higher-value work, Shipley added. “It turns out AI is good at high-value work,” she said, such as “gathering and analyzing data, making predictions and recommendations, [and] prescribing and executing actions.”

But for organizations contending with siloed data, escalating costs, and uncertain ROI, getting to that point might take some doing.