How AI is making self-appraisals better and less painful — for both workers and managers

• AI tools are increasingly used in self-appraisals to aid workers' evaluations of their performance.
• They can help address biases and subjectivity in traditional performance reviews.
• Human oversight is still crucial, especially due to privacy concerns, researchers say.

Performance reviews are not usually something many people look forward to.

Evaluating ourselves is hard, and it can feel like a massive effort to sift through a year's worth of work to find the highlights.

Women can find it particularly difficult to advocate for themselves and emphasize their achievements — and this is where AI might come in handy.

It's already proving popular. According to Microsoft's 2024 Work Trend Index, 75% of global knowledge workers, such as architects, engineers, scientists, and lawyers, are using AI in some form.

When Oracle implemented an AI performance review in 2023 and asked employees if they liked it, 89% said they were willing to be a first adopter.

"That shows how much there is a need and how much people tend to believe in this," Triparna de Vreede, an assistant professor in the school of information systems and management at the University of South Florida's Muma College of Business, whose research focuses on human-AI Interaction and workplace well-being, told BI.

Traditional reviews have downsides

According to de Vreede, performance reviews are "usually a subjective exercise" — one with a lot of "power play" between managers and subordinates.

Employees may think they're doing well, but they're not always aware of how that fits with the company's strategic direction or how their work contributes to the bottom line.

There's also a recency bias, where a lot of the smaller things "sort of get lost in the process," said de Vreede.

"If you have done really great things, and then screwed up once in the past month, that bitterness sort of overlies everything else," she said. "So these are the things that our traditional performance evaluations lack."

AI performance reviews can help with articulation and form statements that align with the company's business strategy and ideology.

They also help managers give better, unbiased feedback, which "creates a more standardized response towards all employees," de Vreede said. "So they don't feel like, 'oh, he likes you, that's why he gave you a raise'."

An objective listener is easier to open up to

De Vreede said that when using an AI performance review, people also tend to speak more. That's because they believe the AI will judge their accomplishments more fairly than their manager.

That's not necessarily because the manager-employee relationship is bad. The best managers cannot analyze their direct reports' emails, documents, and project management tools the way AI can.

People are often unreliable narrators. Dorothy Leidner, a professor of business and AI ethics at the University of Virginia, found this when working on a case study with Johnson & Johnson.

The company wanted to figure out what digital skills the workforce had, but found that challenging.

"It turns out people are not great at estimating their own skills, and some people are overconfident, and then some people are underconfident," Leidner told BI. "So they spent a lot of time working with a vendor and designing an AI to use digital traces and HR data on employees to be able to infer their digital skills."

Women can find reviews challenging

Self-evaluations can be difficult for women, both internally and externally.

"Women are consistently called more emotional and abrasive, whereas men are called more assertive," de Vreede said.

An analysis by the management and HR platform Textio of 23,000 performance reviews across 250 US workplaces last year found that 38% of language used in the performance reviews of high-performing women contained exaggerations, clichés, and fixed-mindset labels. Just over three-quarters of women surveyed said they'd been described as "emotional," compared with just 11% of men.

Women also tend to downplay their achievements.

A study by the National Bureau of Economic Research in 2022 found that women, on average, ranked their performance out of 100 as 46, while men gave themselves an average score of 61.

Meanwhile, women often pick up the slack in the workplace, taking on invisible tasks that pile up and lead to burnout.

"Now imagine if I have an AI that is keeping track of all my achievements," de Vreede said. "It allows us to say, 'oh, this is how we are really making a difference,' without having our imposter syndrome rising up."

People still need to be 'in the loop'

Several companies have already created AI tools for performance reviews including Oracle and Textio, whose AI is used by 25% of Fortune 500 companies to analyze conversations and give feedback.

There is still some way to go before AI self-evaluations are a perfect solution, de Vreede said.

Privacy concerns remain, and companies need to have very clear policies about what is acceptable AI assistance, as well as be transparent about what personal information and work the AI tools can access.

De Vreede said humans need to be "in the loop" so that softer skills, such as communication, warmth, and likeability, are still factored in.

"AI can't understand it on its own, but AI can be trained to ask those questions so that the person can reflect and write about it," she said. "Then you are truly capturing the human as a whole instead of just a workplace creature."

A major concern in de Vreede's view is that relying too much on AI can lead to people failing to self-reflect at all — rendering an evaluation useless.

She said AI can do the hard work, but you are responsible for reviewing the results, editing them, and fine-tuning them. "What comes out as your self-reflection is not just the AI's work, but a collaboration between you and the AI."