So why was this important? “Awareness is the first step,” Cian said. “If retailers aren’t aware of this, they could be wasting a lot of money.”

Here’s how they came to their conclusions:

The researchers divided prospective purchases into two categories: “utilitarian,” or something a customer needs; and “hedonic,” or something a customer wants – in other words, an emotional purchase or “a joy to the senses,” as Cian puts it.

Cian and Longoni found that customers willingly followed computer algorithms – the robot recommendations – for products they needed, like garden tools. But they resisted computer suggestions for things they wanted, like a bottle of wine. For pleasure purchases, customers craved a human recommendation.

The findings are key for retailers that increasingly rely on artificial intelligence – or AI – to produce more accurate and cost-effective marketing strategies.

The research also uncovered an interesting quirk: It turns out that robot recommenders are just as good as humans at guiding emotional purchases, like clothes and accessories, but customers don’t believe that.

“Indeed,” the researchers wrote, “AI selects flower arrangements for 1-800-FLOWERS and creates new flavors for food companies such as McCormick, Starbucks and Coca-Cola.” Yet when customers are told a computer made recommendations for an emotional purchase, they shun the suggestions.

The workaround, Cian and his colleague uncovered, is that customers will still follow the cost-effective computer algorithms for things they want, “as long as a human is somewhere in the process.” The researchers found that human-AI partnerships – an “augmented intelligence” model in which “AI supports rather than replaces a human” – may be the best of both worlds.

One of their studies showed how this works:

Participants were informed that one set of chocolate bars was created from a computer-generated, or AI, recipe. The other bars were largely crafted by a computer recipe, but a human chocolatier oversaw the process.

Consumers who wanted the bars to be healthier (a utilitarian or logical decision), nibbled on both but ultimately picked the bars formulated by computer. Those who simply wanted delicious chocolate (a hedonic or pleasure choice) gravitated to the bars involving the chocolatier.  

Both chocolates were the same.

“These results are important for practitioners managing relatively more hedonic products or services,” the paper notes. “Practitioners could leverage our results and utilize AI systems to generate an initial recommendation on which a human then ‘signs off.’”