If Peter Cappelli, the George W. Taylor Professor of Management at the Wharton School, feels familiar with the current frenzy surrounding artificial intelligence, it's because he's seen the movie before. He points to the period from 2015 to 2017, when major consultancies and the World Economic Forum confidently predicted that driverless trucks would eliminate truck drivers within a few years.
“I didn't have to think very long to realize that it actually didn't make sense,” Cappelli said. luck via Zoom from my home in Philadelphia.
“It didn't take me very long to think about driverless trucks and what happens when you need gas. Or what if you have to stop and make a delivery? And if you have to have an employee on board, of course that defeats the purpose, right?”
Cappelli, who recently partnered with Accenture to uncover the real-world impact of AI on jobs in a series of podcasts, cautioned against listening too much to companies talking about their books or trying to sell new products.
“When you listen to people who are developing technology, they're just talking about what's possible, and they're not thinking about what's practical.”
In a wide range of conversations, luckCappelli addressed what AI actually does to make it work. luck We previously discussed how remote work is actually very bad for most organizations.
“I mean, people say I’m a contrarian,” Cappelli said. “But I don't think so. It's just that I'm skeptical of things, you know?”
When it was pointed out to him that this was essentially a contrarian position, Cappelli laughed before getting back on topic. “I get nervous when there’s so much hype.”
he spoke luck How his research fits into the broader picture that will define the second half of 2025, after an influential MIT study gained attention that found 95% of generative AI pilots failed to generate meaningful profits. His favorite example was a specific case study of a company that put AI to work to both reduce headcount and improve productivity. It still didn't neatly match predictions (according to Elon Musk and Anthropic's Dario Amodei, for example, the job will soon become an option or even a hobby). “It costs a lot of money to do this,” Cappelli said of his discovery. “And this was a success.”
3x the cost
Mr. Cappelli detailed the findings of a case study in which he participated in the Harvard Business Review about insurance claims processing company Ricoh. This is exactly the type of low-level administrative work that AI is expected to be able to easily automate. But the reality of adoption was a financial shock. The company ended up achieving a 3x performance increase, but the transition wasn't cheap. The company spent a year with a team of six people, including three expensive outside consultants, just to get the system up and running.
“What they first discovered was that a large language model could do this very well, and it cost three times as much for an employee to do it,” Cappelli said. [manually]. Okay, that's not going to work. Cappelli noted that the costs included approximately $500,000 in fees paid by Ricoh to outside consultants.
Even after optimizing the process, Ricoh was still spending about $200,000 per month on AI fees. This was more than the total salary for this task. They were able to reduce the number of employees from 44 to 39, which shows that AI is far from being a mass job killer, he added. His explanation brings to mind the example of self-driving trucks.
“The reason they still need employees is because they have to track a lot of issues, and if the issue comes from AI, it's harder to track down,” he said. The good news is that the Ricoh division's productivity will eventually triple, he added.
“That's the reward, but it's not cheap.” [and] It took a very long time. ”
Ashok Shenoy, vice president of Ricoh USA, said: luck After starting to use AI for “very mundane, repetitive, high-volume tasks,” rather than eliminating human jobs, it “moved into areas where human judgment and experience add the most value.” In the roughly one year since this case study was conducted, Ricoh has successfully applied AI to medium-level, repetitive and time-consuming tasks at scale, and the company expects to be able to use AI agents to achieve partial or full automation of workflows within the next six to 12 months, he said. “Humans participate to resolve missing or unclear information and ensure quality.”
While acknowledging the high costs noted by Cappelli, Shenoy noted that the project broke even in less than a year and that the $200,000 monthly cost is cheaper than the previous operating model. “The transition to AI reduced total costs by an estimated 15% without relying on significant headcount reductions.” Regarding headcount, he said, “This effort was not driven by cost or headcount reductions,” noting that implementing AI requires creating new roles, redesigning existing roles, and repurposing team members to higher-value work. He said staffing levels have remained largely stable as productivity has improved and production volumes have increased, and there have been no further layoffs. “The bigger change has been in the way people spend their time. They are doing less repetitive work and focusing on resolving exceptions, maintaining quality, and serving customers.”
The shaming of performative AI in the boardroom
Cappelli said he found similar dynamics in the Accenture partnership with Mastercard, Royal Bank of Scotland and Jabil. “These are all success stories,” he says, and will increase productivity in the long run. Companies will be able to do more with fewer people, but “it's going to take a long time to get there.” He argued that important things are being underestimated. “But what matters is how much effort it takes to do it.”
And when it comes to layoffs, Cappelli said he didn't see any layoffs, at least in specific departments within each company he looked at. When asked for comment by Fortune, Accenture said it largely agreed with Cappelli's conclusions, pointing to a recent interview with CEO Julie Sweet. luck Editor-in-Chief Alison Shontell.
Cappelli said much of the noise surrounding AI, and the distance between potential and practicality, is driven by what other commentators have called “AI shame.”
Cappelli was not familiar with the term “AI shame,” but said: luck He was “absolutely right” to describe what he had seen. “They're pretending so they can say they're doing something, right?” he said. “So investors like this idea, so there's a lot of pressure to make this happen.”
He cited the findings of an early 2025 Harris poll that found 74% of global CEOs felt they would lose their jobs within two years if they couldn't demonstrate the success of AI, and said that about one-third were implementing AI for performance without really understanding what it meant. The Harris Poll says: “CEOs estimate that more than a third (35%) of their companies’ AI efforts are just ‘AI cleaning’ for optics and reputation, delivering little or no real business value.”
Cappelli explained how the market generally celebrates news of layoffs, and also noted research that shows companies announce “phantom layoffs” that never actually happen because they arbitrage the stock market's positive reaction to news of potential layoffs.
Cappelli predicted there will be a “slow learning curve” and CFOs will start to realize “this is a very expensive thing to implement.” The problem, Cappelli said, is that U.S. executives are “coddled” and increasingly reluctant to make organizational change efforts.
“[Employers] I think it should be free. It should be cheap. “You just have to hang the roof, and the right people will show up quickly,” he says. In his opinion, true AI success will require “old-school HR” work: mapping workflows, dividing work into tasks, and having employees collaborate with AI “agents” to refine prompts.
“You can't do it beyond your employees, because they really know how their jobs are done,” Cappelli said. The professor said he was dismayed by what was happening with most executives, saying they were largely “avoiding” the issue of taking this technology seriously.
“They don't see it as an organizational change issue or a big problem,” he says. “They're just adding stress to everyone and hoping it will somehow resolve itself.”
