If the present frenzy over synthetic intelligence feels acquainted to Peter Cappelli, the George W. Taylor professor of administration on the Wharton College, it’s as a result of he’s seen this film earlier than. He factors to the interval between 2015 and 2017, when main consultancies and the World Financial Discussion board confidently predicted that driverless vehicles would get rid of truck drivers inside a couple of years.
“You didn’t must assume very lengthy to appreciate that simply wasn’t going to make sense in apply,” Cappelli advised Fortune on Zoom from his house in Philadelphia.
“You didn’t must assume very lengthy about driverless vehicles to consider, okay, what occurs after they want fuel? You already know? Or what occurs in the event that they must cease and make a supply? And in the event that they must have an worker sitting with them, in fact it defeats the aim, proper?”
Cappelli, who not too long ago partnered with Accenture on a collection of podcasts to unravel what AI is definitely doing to jobs, warned towards listening too intently to the businesses which can be speaking their e book, or making an attempt to promote you on their new merchandise.
“If you happen to’re listening to the individuals who make the know-how, they’re telling you what’s attainable, they usually’re not fascinated by what’s sensible.”
Over the course of a wide-ranging dialog with Fortune, Cappelli tackled what AI is absolutely doing to work, very similar to he talked to Fortune beforehand about how distant work is, really, fairly unhealthy for many organizations.
“I imply, folks say I’m a contrarian,” Cappelli stated, “however I don’t assume so, a lot as I simply am skeptical about stuff, you understand?”
When identified this was an inherently contrarian place, Cappelli laughed, earlier than returning to the primary level. “I simply get nervous with hype.”
He talked to Fortune about how his analysis suits into the broader image that outlined the again half of 2025, after the influential MIT examine that caught the attention on 95% of generative AI pilots failing to generate any significant return. His favourite instance was a selected case examine on an organization that really made AI work, each reducing headcount and boosting productiveness. It nonetheless didn’t match neatly with predictions (say, from Elon Musk or Anthropic’s Dario Amodei, that work will quickly be optionally available, or perhaps a passion). “It’s vastly costly to do that,” Cappelli stated about his findings. “And this was a hit.”
3 times the associated fee
Cappelli detailed the findings of a case examine that he participated in, printed within the Harvard Enterprise Evaluation, on Ricoh, an insurance coverage claims processor: the precise kind of low-level administrative work that AI is meant to automate simply. The fact of adoption, nevertheless, was a monetary shock. Whereas the corporate finally achieved 3 times the efficiency, the transition was something however low cost. The agency spent a yr with a crew of six, three of whom had been costly exterior consultants, simply to get the system working.
“The very first thing they found,” Capelli stated, “is giant language fashions may do that fairly effectively — at 3 times the price of their workers doing it [manually]. Okay, in order that’s not going to work.” Cappelli identified that the prices included Ricoh paying roughly $500,000 in charges to exterior consultants.
Even after optimizing the method, Ricoh was nonetheless spending about $200,000 a month on AI charges—greater than their whole payroll for the duty had been. They had been in a position to lower their headcount from 44 to 39, he added, exhibiting simply how removed from being a large job killer AI is in apply. His clarification remembers his self-driving truck instance.
“The explanation they nonetheless want workers is that numerous issues must be chased down, they usually’re more durable to chase down if they arrive off of AI,” he stated. The excellent news, he added, is that this Ricoh division will finally be 3 times as productive.
“In order that’s the payoff, nevertheless it’s not low cost [and] it took a hell of a very long time to do.”
Ashok Shenoy, VP of Ricoh USA, advised Fortune that, after beginning to use AI for “very routine, repetitive, high-volume duties,” work for people didn’t disappear, however “shifted towards areas the place human judgment and expertise add probably the most worth.” Within the yr or so because the case examine was performed, he famous that Ricoh has efficiently utilized AI to mid-level, repetitive, time-consuming duties at scale, and expects to make use of AI brokers to attain partial or full workflow automation throughout the subsequent six to 12 months, “with a human-in-the-loop to resolve lacking or unclear data and guarantee high quality.”
Whereas acknowledging the big-ticket prices highlighted by Cappelli, Shenoy famous that this venture reached break-even in lower than a yr, and it’s $200,000 month-to-month prices are inexpensive than the earlier working mannequin. “The shift to AI delivered an estimated 15% whole price discount, regardless that it didn’t depend on vital labor cuts.” Relating to headcount, he stated “this train was not pushed by price or headcount discount,” and AI implementation requires creating new roles, redesigning current ones, and repurposing crew members towards higher-value work. He stated there haven’t been additional job cuts, both, with staffing ranges largely stabilizing as productiveness elevated and volumes grew. “The larger change was in how folks spent their time. They’re doing much less repetitive work and are extra centered on resolving exceptions, sustaining high quality and serving prospects.”
Performative AI disgrace within the boardroom
Cappelli stated he discovered related dynamics in his partnership with Accenture, which checked out Mastercard, Royal Financial institution of Scotland, and Jabil. “These are all success tales,” he stated, and in the long term, they are going to see productiveness will go up. Corporations will be capable to do extra with fewer folks however “it’ll take a protracted whereas to get there.” He argued that one thing essential is being underestimated. “The important thing factor, although, is simply how a lot work is concerned in doing it.”
Additionally, concerning headcount reductions, Cappelli stated that at the least within the areas that he researched, which had been particular models inside every firm, he didn’t see any job cuts in any way. When contacted for remark by Fortune, Accenture stated it largely agrees with Cappelli’s conclusions, and referred again to CEO Julie Candy’s current interview with Fortune Editor-in-Chief Alyson Shontell.
Based on Cappelli, a lot of the noise round AI—and the gap between what’s attainable and what’s sensible—is pushed by what different commentators have known as “AI disgrace.”
Cappelli wasn’t aware of the “AI disgrace” phrase, however advised Fortune it was “completely proper” in describing what he’s seen. “They’re pretending to allow them to say they’re doing one thing, proper?” he stated. “So the strain is simply huge on them to attempt to make these things work, as a result of the buyers love the concept.”
The professor cited the Harris Ballot’s discovering in early 2025 that 74% of CEOs globally felt they’d lose their job in two years in the event that they couldn’t reveal AI success, and roughly a 3rd stated they had been performatively adopting AI with out actually understanding what it might entail. As The Harris Ballot put it: “CEOs estimate that over a 3rd (35%) of their AI initiatives quantity to mere ‘AI washing’ for optics and status, however providing little to no actual enterprise worth in any respect.”
Cappelli described how markets usually rejoice information of layoffs, and even cited analysis that “phantom layoffs” get introduced by firms that by no means really happen, as a result of firms are arbitraging the optimistic stock-market response to the information of a possible layoff.
Cappelli predicted a “gradual studying curve” will happen, through which CFOs will begin realizing “that is super-expensive stuff to place in place.” The issue, in response to Cappelli, is that U.S. administration has change into “spoiled” and more and more averse to the arduous work of organizational change.
“[Employers] assume it needs to be free. It needs to be low cost. It is best to simply be capable to dangle a shingle out, and the suitable folks will simply present up,” he says. Actual AI success, in his opinion, would require “old school human assets” work: mapping workflows, breaking down jobs into duties, and having workers work alongside AI “brokers” to refine prompts.
“You may’t do it excessive of workers, as a result of the workers actually do know the way their job is completed,” Cappelli stated. The professor was withering about what he sees taking place in most C-suites, saying they’re largely “ducking” the issue of actually grappling with this know-how.
“They’re not seeing it as a company change drawback and an enormous one,” he stated. “They’re simply stressing everyone out and, you understand, hoping that it by some means works itself out.”
