Delegating Decision-Making to the Machine: Experimental Evidence from Health Insurance

With the proliferation of on-line shopping tools and the advancement of predictive algorithms, personalized decision-making support software for consumers - especially in markets for household finances - is becoming commonplace. Does delegating consumer decisions to algorithms affect consumer choices and market efficiency? We present the results of a randomized controlled trial in which we offered (elderly) consumers a decision-making support software for choosing among pharmaceutical insurance plans. We find that algorithmic “expert” recommendation significantly alters consumer choices (plan switching rate increases by 8pp relative to 28% baseline rate). At the same time, personalized, but passive, informational treatment has little effect on switching. We find that selection into who uses the support tool is quantitatively large. Consumers that are more likely to switch their insurance plans are much more likely to take up our intervention. We find similar patterns for other measures of choice behavior. We use a discrete choice model of consumer decision-making to analyze the mechanisms that underlie the aggregate treatment effects and to estimate how the algorithmic decision-making support affects consumer welfare.