NewtonX recently conducted a survey of 1,500 American full-time, salaried employees in office jobs with the intent of understanding what employees want in exchange for improved productivity, and also conducted a sister survey with 200 senior level executives at these same companies, in order to understand the disconnect between employees and employers, when it comes to improving output (you can read an excerpt of the productivity findings here). One interesting finding from this survey was that only 9% of employees and 22% of senior executives believe that their performance management processes are effective, and in particular view reviews and feedback as painful. Employees and executives alike believe that performance reviews are rife with bias, unclear objectives, and conflicting definitions of success.
AI provides an appealing alternative. IBM claims that its AI, Watson, can predict future employee performance with 96 percent accuracy. In general, AI has a much better memory and ability to measure metrics against one another (such as an employee’s goals/standards to their actual performance) — an AI isn’t likely to forget about your strong first quarter, judge your lipstick, or read into your lunch habits. Furthermore, AI can give real-time feedback, negating the need for retroactive chastising months after the fact.
So why aren’t we using AI for performance reviews? NewtonX consulted AI experts as well as the executive and employee survey to identify the benefits and drawbacks of letting AI do performance management.
Would AI solve bias in performance reviews, or exacerbate it?
One of the appeals of AI for performance review is that it is likely to be significantly less biased than humans are. Not only can AI correct for racial and gender bias, but it also is not susceptible to performance-review-specific biases, such as recency bias (where actions performed recently are given more weight than actions that occurred say, 11 months ago for a yearly assessment). Similarly, AI can control for contrast bias, which occurs when a manager compares an employee’s performance to their peers rather than to objective measures of success. This bias can be particularly pervasive in growing companies — perhaps the entire sales team met their goals, but an evaluator may be inclined to give the least successful representative a worse performance review, even though they objectively performed to standard.
That said, AI algorithms are only as good as the training data they are fed. While many algorithms today can detect and flag bias based on gender or race, data needs to be weighted and examined to ensure that other types of human bias don’t teach the AI to adopt the same biases. It’s an old saying in computer science: garbage in, garbage out. AI learns from the data that humans feed it, so in order for it to be better than us when it comes to bias, the AI will need to taught through unbiased data. AI doesn’t suffer from logic fallibility as long as the data its given doesn’t either.
A numbers game: why the market for AI- powered performance review is skyrocketing
AI-powered assessment can happen in real-time based on targets, quotas, and progress toward each. Because it can react in real-time, it can also give positive reinforcement, incentives, and alerts when performance is slipping. It can even give clear feedback on areas that need improvement to bring the employee up to their targets. The real-time nature of AI feedback can save companies millions in time and productivity: IBM, for instance, has 380,000 employees worldwide; giving just two hours of feedback once per year for each employee is a colossal time suck, especially considering that yearly reviews tend not to impact future performance.
It’s no wonder, then, that IBM is not the only company moving toward AI-powered performance review. As early as 2011, yearly performance reviews began to fall out of vogue in favor of continuous feedback. Companies including Cargill, Adobe, and Accenture abandoned the old GE model of “pass or perish” yearly reviews, and by some measures, cutting the yearly review in favor of continuous feedback has reduced turnover by 30% (which also equals millions of dollars in recruiting and lost productivity for onboarding/ disengaged employees).
Today, the HR and workforce management software market has reached $12B, and is projected to grow 25% year over year for the next three years. Companies including BambooHR, Nexus AI (which in addition to automated performance reviews also algorithmically determines which employees should work together on given projects), and HR behemoths like Zenefits have developed products that give employees and managers continuous feedback, not only on performance but also on context-specific performance — e.g. who works best on what projects and with whom.
That said, even executives who already use or are planning to use AI-powered performance management tools say that they combine these tools with personalized oversight. “Nobody is getting fired by an AI,” declared a VP at a supply and logistics company that uses AI-performance management software. As in so many other AI application areas, the AI is there to augment and improve human efficiency and productivity; not to replace it altogether.