The paper “Strategic Responses to Algorithmic Recommendations: Evidence from Hotel Pricing” by Daniel Garcia (U Vienna), Juha Tolvanen (U Tor Vergata Rome), and Alexander K. Wagner (PLUS) is now available in Management Science, a top journal for research in Business, Economics and Management. The paper theoretically and empirically studies the strategic interaction between an algorithmic AI advisor that recommends prices and a human manager who sets prices based on the algorithm’s price recommendations in the context of hotel room pricing.

We find that there are large price-setting frictions, as human managers find it difficult to adjust hotel room prices due to the sheer volume of price recommendations provided by the recommendation algorithm, suggesting that human managers face high adjustment costs. Using a large dataset of hotel prices with algorithmic price recommendations, we show that the AI advisor has an incentive to exaggerate its price recommendations (strategically bias its recommendations) to induce the human manager to implement its recommendations. This leads to suboptimal pricing of hotel rooms by human managers. Our structural estimates show that fully delegating pricing to the AI advisor, and thus neglecting the information that human managers have about hotel room prices, would increase hotel revenues.

More generally, the novel strategic bias in AI advisor recommendations we present is likely to be ubiquitous in settings where humans ‘in the loop’ have access to algorithmic default or status quo options, and may affect not only price recommendations from AI advisors in revenue management, but also treatment recommendations from doctors, probation recommendations from judges, and so on.

Read the full article  here.