Workers using AI tools are starting to sound all too similar. How can firms assess the talent that can make or break their future?
Perfectly pleasant
Scroll to read
VIEW THE
PODCAST
READ THE
FULL MAGAZINE
he irony is not lost on Rohit Agarwal.
He co-founded a technology staffing firm, Zenius, that connects software developers with small and mid-sized customers worldwide who want to keep their own full-time workforces lean. One of his big selling points is that Zenius takes pesky HR problems, such as evaluating how well each of those software developers performs, off of his clients’ hands.
But Agarwal has run into a new problem: He’s having trouble determining which of his own developers are actually good. Managers used to be able to judge a developer’s performance almost entirely on their speed. The faster someone could build properly working software, the better a developer they were. Now, thanks to AI, almost anyone can code quickly and competently—and Agarwal can no longer determine which of his developers might be able to tackle far tougher projects in the future. Since his teams work remotely, clients don’t have a great deal of feedback to offer either. He has realized he has to revamp his entire employee-evaluation process. “The old performance signals are less reliable,” he says.
AI has created a battle for firms trying to assess talent accurately.
The
Problem
Firms excel by weeding out low performers and moving up stars.
Why It
Matters
Upend the performance-review process and allow more one-on-one evaluations.
The
Solution
T
”
”
By Russell Pearlman /
Photo Illustrations by Tim Ames
Ranking Employees: Looking Back
AI may be bringing the effort to the forefront, but companies have
been trying to evaluate worker performance for a century.
The hope, of course, has been that AI could somewhat rectify all this. Both its remarkable tracking ability and data storage might create some handy objective measuring tools. And initially, that was the hope for corporate recruiting—that AI tools could, at scale, judge candidates online. The only problem: Candidates got hold of
the tech tool, too, and at a level no one could have imagined. In the last two years, recruiters have been deluged with tens of thousands of AI-polished résumés and cover letters. Every one of them, it seems, makes the candidate sound ideal for the job... and none stands out. Now, as the issue of sameness moves into the actual workplace, firms are on the hunt for solutions. “AI has not made everyone an average performer, but it has made average work look better,” says Pennington.
AI:
Compressing Skill Levels
AI evangelists tout how the technology can give every worker superpowers. And it’s true; AI helps produce a lot of stuff. Need to build a complex financial model? Presto—done in an hour. Need a ten-page marketing presentation? Poof—ready in 20 minutes. Need to send out 50 emails to clients about a new product? Whoosh—written and sent in five minutes. All of this output is pleasantly perfect, as a cynical manager might say, but experts fear the technology may make it much harder to determine what kind of employee did the work: Was it a toxic slacker who just knows a couple of good AI prompts? A highly curious achiever who yearns for tougher work? Someone in between? Indeed, research has showed that below-average employees using AI improved their performance at a much higher rate than high performers using it. Increasingly, leaders are facing a nightmarish scenario in which everyone can produce the same stuff, or at least stuff that all looks and sounds the same. “I would no longer evaluate top performance based on how polished the final product looks,” says Melissa Pennington, a longtime HR executive who founded Future of
HR Academy.
to innovating, can wither. It’s a problem that hasn’t yet appeared in real life, but it’s already showing up in academic research. In one experiment, Harvard University researchers asked several hundred consultants to use the popular AI model ChatGPT to solve a few problems. The consultants rated as below-average performers did 43 percent better on tasks helped by AI‚ while those rated above average performed only 17 percent better. In other words, AI raised the “floor” in the survey but didn’t up the game of the top consultants much. What if even top performers don’t pursue answers beyond the ones AI gives them? The same skill-compression effect also shows up in a recent National Bureau of Economic Research study of 5,000 customer-service workers: When they used AI assistants, low-skilled workers saw a 35 percent gain in productivity, but high-skilled workers saw almost none. “I’m happy to see lower-quality work improve, but sameness comes at a cost,” says Bob Hutchins, an adjunct faculty member of Lipscomb University who researches performance management.
To be sure, some managers look at the issue more positively: They are thrilled that AI has created a higher floor for work quality. After all, higher productivity from direct reports will probably look good on a manager’s own performance review. But experts warn that this attitude could be short-sighted. Decades of research show modern organizations can’t excel without at least a sliver of top-performing employees. In the years leading up to the introduction of large-language AI models, a tiny number of corporate superstars produced a massive amount of total value, while the vast majority of employees performed at substantially below-average levels, says Ernest O’Boyle, a management professor at Indiana University whose 2012 paper “The Best and the Rest” described the phenomenon. “AI isn’t likely going to change that dynamic,” he says.
The first serious challenge for many organizations, experts believe, will be determining how to evaluate junior-level employees. In many cases, the current crop of young workers hasn’t shown much interest in climbing the corporate ladder (according to many surveys, about 70 percent of them would rather spend their entire careers as individual contributors than move into management). But being able to evaluate the performance of this group on traditional junior-level assignments—drafting summaries, conducting raw research, building financial models—could help an organization separate those who are truly talented and ambitious from their peers who’d rather just go with the flow. The only snag: AI is now handling those foundational tasks.
But the most troubling question is this: What if AI’s “pleasantly perfect” output becomes the standard for all employees, and truly inspired work all but disappears? Experts worry that when workers rely on AI, their critical skills, from problem-solving
Ultimately, the seemingly simple task of assessment became the stuff of corporate legend, with one Fortune 500 company last decade revealing it was spending an astounding 2 million hours per year on annual performance reviews (the company eventually scrapped
its yearly evaluations). Many other firms also opted to ditch annual ratings, asking managers to evaluate their direct reports in real time. “Almost nobody we speak to in the corporate world seems satisfied with their performance-management process—and absolutely
nobody knows what to do about it,” says Lesley Uren, CEO of Korn Ferry Consulting.
Decades later, as manufacturing started to give way to a more consumer- and service-
oriented economy, managers shifted their focus from physical activities to accomplishing
goals. The shift made sense, of course. Most organizations haven’t thrived over the last century just because they made stuff faster; they’ve succeeded because their employees came up with new ideas, provided excellent customer service, motivated coworkers, or achieved an assortment of other objectives that aren’t solely measured in terms of time or money.
But finding the right metrics to rank workers this way didn’t prove easy, and neither did knowing what to do with performance reviews. Some firms took rankings to an extreme, forcing bosses to summarily fire employees who underperformed in any given year. Even firms that didn’t use the so-called “rank and yank” evaluations found themselves turning to metrics that weren’t objective and allowing the biases of bosses to rule the day. In many cases, managers discovered they didn’t have a great view into the work all of their direct reports did or would overvalue employees simply because they were good communicators. And good luck to the manager asked to rank people with different roles.
When it first emerged as a business practice at the start of the 20th century,
performance management seemed straightforward. In 1910, firms began asking
factory managers to rank, in a systematic way, how fast employees could operate machinery or assemble products. For firms, the appeal of distinguishing hard
workers and talented performers from those going through the motions was
too hard to resist. By 1920, 60 percent of large US factories used some form
of ranking system.
The Early Days of Employee Reviews
No one suggests that organizations should stop using AI because it’s making performance management tougher. The technology has too much potential—and, in an increasing number of cases, proven ability—to improve quality and productivity. Instead, experts believe that evaluation processes need a refresh. One suggestion for managers: Make employees defend their decision-making either during or after they finish an assignment. Universities have used this process successfully for centuries in conducting thesis defenses, says Hutchins. Today, asking a company’s employees how they came to their conclusions can help determine how much they relied on their own judgment and how much they relied on AI.
Assessments, Up Close and Personal
Other experts suggest that managers divide up performance management into two buckets: “stuff AI can help with” and “everything else.” And, at least for the foreseeable future, there’s plenty of stuff that goes into the “everything else” bucket. Demonstrating ethical judgment, mentoring junior staffers, influencing stakeholders, displaying empathy, and motivating coworkers are human skills that AI can’t itself accomplish. To be sure, implementing both suggestions might require firms to hire more managers, when many have been reducing their ranks. But many experts say it’s probably worth it.
Whether by hiring more managers or changing how they evaluate performers, some executives are already adapting. For instance, AI hasn’t changed Stephanie Heathman’s performance-evaluation metrics. The founder of consultancy HR Innovator Group still assesses how well her team generates additional business and uncovers waste for its clients. But after seeing the early results of AI-enabled work, she anticipated the performance-management headaches. So, she instituted AI training to get everyone to a baseline skill level and, importantly, she has started asking different questions during one-on-one meetings. She wants to know how the firm’s employees are solving problems in the numerous situations in which AI can’t help them. “AI should not just get people to the average,” she says. “It should help them accelerate.”
look better.
average work
AI has made
WATCH
PODCAST
Talent Trends 2027:
Skills-Shifting
Related Insights
READ FULL
ISSUE
The AI Productivity Problem
Test Your Organization’s
AI Readiness
Organizational psychologists have coined a jargony name for all this: “signal pollution.” Separating the good employees from the mediocre ones—and sniffing out potential stars—is already a massive problem for recruiters. Job candidates, regardless of their actual talent, can use AI to produce polished, perfect-seeming résumés and send them off to hundreds of potential employers at once. But as the new technology winds its way through an organization’s numerous workflows, experts worry that evaluating who’s actually a standout will become a much bigger problem. And the risks are far from minor; in fact, experts say they are nothing less than a company’s ability to innovate and grow: Numerous studies show that a tiny number of superstar employees can account for 80 percent or more of an organization’s value.
”
based on how polished the final product looks.
I would no longer evaluate top performance
”
”
management process.
satisfied with THEIR performance-
Almost nobody SEEMS
”
”
rectify this.
AI could somehow
The hope today is that
”
The tech’s ability to polish work can make anyone look good, creating challenges for firms seeking top performers.
2020s (the AI Era)
It’s not just about the money. Employees get ranked on customer satisfaction, team building, diversity, and other metrics.
2000s–2010s
The “Rank and Yank” Era: Each year’s top 20 percent of performers are rewarded and its bottom 10 percent of performers are fired.
1960s–1990s
Peter Drucker’s The Practice of Management shifts the focus of evaluations from completing physical tasks
to accomplishing goals.
1950s
Employees are ranked on clear production quotas and offered bonuses if they can top them.
1920s–1930s
Frederick Winslow Taylor’s The Principles of Scientific Management introduces the idea that there’s “one best way” for factory employees
to do their jobs.
1910s
Rules of thumb—unwritten, imprecise estimates of what workers should be able to do—are the norm.
19th Century
The tech’s ability to polish work can make anyone look good, creating challenges for firms seeking top performers.
2020s (the AI Era)
It’s not just about the money. Employees get ranked on customer satisfaction, team building, diversity, and other metrics.
2000s–2010s
The “Rank and Yank” Era: Each year’s top 20 percent of performers are rewarded and its bottom 10 percent of performers are fired.
1960s–1990s
Peter Drucker’s The Practice of Management shifts the focus of evaluations from completing physical tasks
to accomplishing goals.
1950s
Employees are ranked on clear production quotas and offered bonuses if they can top them.
1920's–1930's
Frederick Winslow Taylor’s The Principles of Scientific Management introduces the idea that there’s “one best way” for factory employees
to do their jobs.
1910's
Rules of thumb—unwritten, imprecise estimates of what workers should be able to do—are the norm.
19th Century
