Performance reviews are stressful for everyone involved, with 22% even admitting to crying afterward. So, as artificial intelligence (AI) continues to gain more ground in improving the HR space, why not use it for performance reviews?
It seems simple enough. An unbiased party analyzes the data to report how well an employee performs and how they could improve. Employees don’t have to worry about the dreaded performance meeting or getting judged by AI. What could go wrong?
Unfortunately, there are a few risks things to consider.
Loss of Human Empathy in Evaluations
Understanding Context
AI cannot evaluate the whole worker. Unless somehow prompted, it won’t understand how one employee’s productivity was low that quarter because they had an ill family member, or how another employee was facing burnout due to covering the projects of multiple colleagues. By simply looking at the data available, AI cannot conceptualize the bigger picture encompassing every factor affecting an employee’s work performance. Meanwhile, a manager would likely already be aware of these circumstances. If not, the performance review would then offer a space for the employee to share what is affecting their work.
Finding Empathy
Your employees are more than numbers, but that’s difficult for your AI system to understand. Even if the AI is given context, it can still struggle or completely ignore the motivations and emotions behind a worker’s performance. An AI’s response to a less-than-perfect worker can be cold and detached with little concern for how feedback is delivered. Remember that 22% from earlier? Removing empathy from the already stressful process is a perfect recipe for disaster.
Building Trust
These performance reviews are not just an avenue for employees to provide the necessary context about their work—they are also needed to help build trust. The worker doesn’t just need empathy from their manager. They also need someone who will give them constructive feedback with the motivation of supporting the worker to improve in their career. Meanwhile, AI might only be concerned with how much of the job is getting done and how well. By leaving out the manager, AI risks alienating workers by causing them to not feel valued enough for a human conversation on their performance.
Accuracy vs. Algorithmic Bias
Is It Accurate?
If the data entered into your AI software is accurate, then the resulting output should be accurate as well. Right?
Not quite.
AI can analyze several performance metrics, including productivity data, error rates, attendance data, response time, and customer satisfaction. While these are all important for company and employee success, they do not provide the full picture behind an employee’s performance. Soft skills such as emotional intelligence, adaptability, flexibility, creativity, teamwork, and initiative are just as important as quantitative metrics. Yet, AI struggles to evaluate just how strongly these characteristics are present within an employee’s behavior. Employees who exhibit these soft skills should be acknowledged, and those who are struggling in this area should receive extra support. If the task is completely left to the AI software, the resulting performance review is at risk of being inaccurate, which could then negatively affect how the employee sees their efforts and even themselves.
So maybe the performance data you input is accurate, but it is likely incomplete.
The Risk of Algorithmic Bias
While AI might seem like a good tool to evaluate without bias, the truth is a bit more complicated.
Just like every other task it takes on, AI must be trained on performance reviews beforehand. However, that training data, even if it is unintentional, could very well be biased.
Performance reviews already struggle with a bias problem without AI in the mix. A report on performance feedback from Textio found that white employees are usually described as easier to work with, and men are more likely to be called ambitious. Meanwhile, Black employees receive less feedback overall, and 26% say that feedback is without any direction. Women also end up feeling more underappreciated than men—approximately 1.3 times more often. If that data reflects existing prejudices—whether related to race, gender, or other factors—AI can unintentionally perpetuate and even exacerbate these biases.
Next Steps
AI can make numerous HR processes more simplified and streamlined, but performance reviews require a balance. After analyzing quantitative metrics, AI must be supplemented with human empathy to account for soft skills and personal challenges. By using AI as a support tool rather than a replacement, organizations can foster more accurate, inclusive, and constructive evaluations that empower employee growth.
