Recommending What to Search: Sales Volume and Consumption Diversity Effects of a Query Recommender System

We consider the sales volume and consumption diversity effects of a query recommender system (QRS). Collaborating with the leading mobile, on-demand food take-out app in Asia, we design and implement a field experiment wherein we randomly assign some users' access to a QRS within their text-based search interface. We show that the QRS drives an approximate 1-2% increase in user order volumes over a 30-day period. Further, we show that the QRS also increases diversity of consumption, not only at the individual level, but also at the level of the market. We provide direct evidence of the QRS's diversity-enhancing impact, showing that treated users begin to employ systematically more generic and shorter search queries, exposing them to a broader array of merchants and products. Finally, we demonstrate that the the value of the QRS depends directly upon the complementary auto-complete feature. We show that the users who respond most strongly to the treatment are those who relied on the auto-complete to a greater degree prior to the introduction of the QRS. Further, we show that treated users increase their reliance on the auto-complete feature relative to users in control, such that treated users reduce their query volumes when not using auto-complete, and increase their query volumes when using auto-complete. Lastly, our findings help to inform platform managers about the impacts of incorporating a QRS both in terms of changes in user search behavior and, ultimately, the economic value that it generates within a platform.