Does Search Ranking Matter?

An A/B test analysis of Expedia's hotel search ranking

Suhail Mulla · Georgia Tech MS Analytics

10M+ Hotel impressions analyzed
-2.0% Clicks per session (algorithm vs random)
93.9% vs 13.0% Booking rate gap (data artifact, not real)

Expedia A/B tested their hotel ranking algorithm against random ordering across 400,000 search sessions. The headline booking rate looked like a 620% lift, but it turned out to be a data construction artifact. The one clean metric, clicks per session, shows a small but real 2% reduction in the algorithm group. Whether that means better relevance or disengagement is something the data alone can't answer.

Bootstrap distribution of clicks per session difference
Bootstrap distribution of the click difference. The entire distribution is below zero, confirming the algorithm group has fewer clicks per session.