Should you be using AI for performance reviews?

Here are four possible scenarios to consider.

[Photos: LinkedIn Sales Solutions/Unsplash; kjpargeter/FreePik]

BY Tomas Chamorro-Premuzic and Ben Dattner

During the last decade, digital innovations have produced a range of recruitment and evaluation tools: now, whenever you first apply for a job, you are less likely to be judged by humans and more likely to be assessed by AI. Before you can even get the opportunity to impress a human interviewer, you will first need to impress the algorithm!

More recently, AI has also been used to assist current employees in doing their jobs and then to help their employers evaluate how well employees are performing in those jobs. In fact, AI adoption is now the norm across knowledge economy jobs, with estimates indicating that at least 70% of people use AI regularly at work (a figure that is probably an underrepresentation of the reality, since much of AI use at work is clandestine and undisclosed) and an increasing number of organizations are using AI to evaluate employee performance.

Meritocratic or Orwellian?

Traditional performance evaluations (often an onerous, annual ritual based on subjective, “noisy,” and unreliable or invalid manager feedback) are indeed being disrupted by algorithms capable of analyzing workflows, communication patterns, and even “relational analytics” (mining the digital footprints of your exchanges with coworkers) in real-time, which critics lament as a form of “surveillance capitalism.”

Get more insights from Tomas Chamorro-Premuzic

Dr. Tomas Chamorro-Premuzic is a professor of organizational psychology at UCL and Columbia University, and the co-founder of DeeperSignals. He has authored 15 books and over 250 scientific articles on the psychology of talent, leadership, AI, and entrepreneurship. 

To be sure, these tools put unprecedented power in the hands of organizations to pursue data-driven management decisions which, at their best, can make workplaces fairer and more meritocratic, but at their worst, seem uncomfortably close to an Orwellian big brother dystopia and can erode trust and morale.

To make sense of AI in performance management, it helps to imagine a simple matrix with four quadrants or scenarios, which echoes the classic negotiation model by Roger Fisher and William Ury on win-win outcomes, as well as decades of behavioral science differentiating integrative from zero-sum approaches to conflict. In one scenario, the company and the employee both win. In another, only the company wins. In a third, employees learn to game the system to their benefit but not to the company’s. And in the worst case, nobody benefits at all.

First scenario: AI helps both the company and the employee. Let’s start with the best quadrant of the matrix. Used well, AI can make feedback fairer and more useful. Anyone who has ever received a vague appraisal knows the problem, and meta-analytic studies show that only 1/3 of feedback is typically useful, 1/3 is useless or irrelevant, and 1/3 actually worsens employees’ performance! Add to this the typical unreliability of performance evaluations, which are usually highly subjective: one manager loves your enthusiasm; another thinks you talk too much; a third simply remembers the last mistake you made; a fourth has no idea who you are, and so on. In other words, performance evaluation has historically been closer to subjective wine tasting than to objective science.

AI, if properly used and validated, can anchor feedback in observable behavior rather than impressions. A sales manager might see which client interactions actually led to repeat business on her sales team. A project leader might learn that delays happen when approvals pile up on his desk. Instead of impatiently waiting for an annual review to learn how their performance may be perceived, employees get real-time feedback and suggestions. The process becomes closer to coaching than judging. This is where the promise of AI is most compelling. It democratizes the collection and distribution of feedback and suggestions. It replaces guesswork with data. It never forgets and it can make employee evaluation performance-driven rather than political.

Artificial Intelligence

Claire's went from tween mall icon to bankrupt — twice?

© Fast Company