The Challenges of Deploying an Algorithmic Pricing Tool: Evidence from Airbnb

Abstract

We study the challenges of deploying an AI pricing tool, Smart Pricing (SP), which was developed by Airbnb to help hosts price their properties. SP was made available for free and was especially meant to benefit those hosts who rarely changed their prices with temporal demand fluctuations. However, SP's adoption rate was low, especially amongst those hosts who rarely changed their prices prior to SP's introduction in the market. We estimate a structural model to (i) identify the extent to which adoption costs, hosts' pessimistic prior beliefs about SP's performance, and sub-optimality of SP contributed to its low adoption rate, and (ii) suggest ways to improve SP's adoption rate that would in turn increase Airbnb's and hosts' profitability. We find that (a) the main reason for SP's low adoption rate is hosts' pessimistic prior beliefs about SP's performance, which is stronger amongst hosts who rarely changed their prices; (b) the benefits of SP for hosts and Airbnb can be increased if Airbnb were to educate the hosts about SP's actual performance; and (c) since Airbnb does not have information on hosts' marginal costs, retraining SP using the structural estimates of marginal costs can substantially increase Airbnb's and hosts' profitability.