Does performance pay boost AI usage?

Advocates of performance-related pay believe that it enhances performance by offering clear, financial incentives to achieve specific goals. It's also likely to have certain other consequences—many of which are unintended.

For instance, research from CU Boulder shows that performance-related pay can demotivate employees, especially if they have a poor boss. This is because they rely on their boss to support them in achieving those stretch goals, and without that support, the goals can feel unachievable.

Similarly, research from Bocconi University found that performance-related pay can harm innovation because it encourages short-term thinking, while also incentivizing employees to collaborate with a smaller pool of colleagues.

Driving AI usage

Might incentive pay also play a role in driving artificial intelligence (AI) usage in organizations? That was the question posed by recent research from Cornell, which looked at whether incentivized pay schemes encourage managers to incorporate AI into their decision-making.

“We examine how compensation contract design and framing of an AI algorithm influence decision makers’ reliance on algorithmic advice and performance in a price estimation task," the researchers explain.

They recruited around 1,500 volunteers and asked them to estimate the price per night of various Airbnb properties in Vienna. The participants received either a fixed sum for completing the assignment, performance pay that incentivized good results, or so-called tournament pay, which rewards pay based on a ranking with one's peers. The volunteers were also given either AI-based advice, human and AI advice, or no advice at all.

"Participants receive no algorithmic price predictions, price predictions with a description focusing on the random forest model of the AI algorithm, or price predictions with a description highlighting the involvement of human experts," the researchers explain.

Earning their pay

The researchers hypothesized that when decisions were incentivized, whether through traditional performance-related pay or via tournament pay, people would be more inclined to try and "earn" their pay by making their own decisions rather than relying on external support.

They wanted to see not only if modern AI-based decision support tools produced a similar outcome, but also whether the way such tools were framed would influence how readily people adopted them.

For instance, some people were told that it was purely AI on its own that was aiding their decision, whereas others were told that it was an AI working with a human expert.

This framing played a huge role, as under the right conditions, people were happy to confound the long-standing aversion to algorithmic involvement and lean on AI for support.

Framing matters

The volunteers had to estimate the price per night for various Airbnb properties. They each had all the relevant information about the property, except the price, with the two AI groups also given a recommendation from an algorithm.

"Both types of incentive contracts lead to a substantially higher reliance on algorithmic advice compared with a fixed payment," the researchers explain. “Performance-based and tournament incentives have similar positive effects on effort duration and positive but statistically not significant effects on performance."

Perhaps most importantly, however, those in both of the incentivized groups were also more likely to rely on the AI than those offered a fixed fee for completing the job. What's more, those who took the AI-based advice also performed better than those who didn't get any help from the algorithm.

The framing also seemed to matter, with participants more inclined to utilize the AI support when they were led to believe that a human was also involved in producing the recommendation.

Supporting adoption

The findings counter the traditional view of algorithmic aversion, with the researchers confident that their study has real implications for how organizations can encourage AI adoption at work, especially at a time when many managers seem to want employees to use the technology.

"There is little evidence of incentives working against the uptake of algorithmic advice for both main incentive types: individual and tournament incentives," the researchers explain. "A simple framing intervention highlighting the involvement of human experts in the development of the AI decision tool can also positively affect attitudes toward algorithmic advice."

So, if organizations want to encourage the adoption of AI among the workforce, implementing performance-related pay could be beneficial, as could ensuring that AI-based recommendations also have some human input. How you motivate people and how you talk about the technology are crucially important.