Martin León Barreto
Companies are constantly making decisions about how to display product information online, and having more consumer data only brings more choices. A retailer selling laptops could emphasize price, screen size, battery life, customer reviews, or any number of other features. But how does the business know which attributes should be highlighted to steer consumers toward products that maximize their satisfaction—and the company’s long-term profits?
Research by Yale’s Jason Abaluck, Chicago Booth’s Giovanni Compiani, and Fan Zhang from the Nova School of Business & Economics suggests a method that retailers can use to help guide their decisions.
For simplicity, research in marketing and economics often assumes that customers are fully informed. However, in real life, shoppers may not know about a feature that they might otherwise value highly. Studies that ignore this possibility when analyzing data on consumer choices will typically underestimate how much consumers care about those features, according to Abaluck, Compiani, and Zhang.
“Models that assume full information may generate wrong conclusions about welfare and—by construction—cannot be used to assess how consumers would respond to an information intervention,” they write.
An added complication is that consumers use different search strategies. Some look at many different options at the same time. Others examine one product in detail and buy it if it fits their needs—but otherwise move on to another item. This makes it challenging to model consumer search in a way that generalizes across different product categories and shopping environments, says Compiani.
To address these problems, the researchers developed an approach to learning consumer preferences that does not assume they have all the information and does not hinge on knowing exactly how consumers search. The innovation behind the method is technical, but in short, instead of measuring how much customer demand shifts in response to changes in a product feature, the model measures how much the responsiveness itself varies between buyers. This approach allows the researchers to isolate the choices of shoppers who behave as if they are fully informed even when they do not investigate every product.
Hidden influence
In a real-world study involving Expedia, shoppers could view a hotel’s pricing, review scores, and star ratings, but they were required to take an extra step to learn its location. Because an unseen attribute can appear to have little influence on choice, a standard approach underestimated how much shoppers valued location. The researchers’ more flexible approach revealed its greater importance.
Retailers can use the method to test whether buyers are actually informed, as well as to forecast responses to new information and to calculate how much shoppers would benefit from learning more about a product or service.
The researchers tested their model’s predictions in two ways. First, they conducted an experiment in which 196 participants chose one e-book from sets of three randomly selected books sold on Amazon Kindle. For each book, people could see a photo of its cover; the title, author, and genre; its rating on the Goodreads website; and the total number of reviews. Each book’s price was also visible, randomly set at $11–$15.
The researchers then varied whether participants could see a randomly assigned discount for each book: For five of the 20 book selections, people could see the discounts without needing to do anything further. For the additional 15 selections, participants had to decide whether to spend a portion of $15 in credits provided to them for each choice. This setup simulated the effort of searching.
People with full information about a product discount should care only about the final price of the book, and the researchers’ findings support this theory. But when the participants had to pay to learn about a discount, they appeared to be less responsive to it than to the initial price. Why? In many cases, they simply didn’t see it. They might have valued a price cut but chose not to spend their credits to learn the amount of a discount, and thus were ultimately unaware of it. Standard models that do not account for this behavior underestimate how much customers care about discounts, the study points out.
In a second test, the researchers applied their model to hotel-choice data from Expedia, looking at purchases made by more than 54,000 travelers from November 2012 through June 2013. Expedia’s site required browsers to click a link to discover a hotel’s location.
The model, using only shopper behavior and without examining the website itself, proved able to figure out that the location was initially hidden. While this was a simple test, it indicated that the researchers’ model could be more successful than standard approaches at valuing features such as hotel location. And it demonstrated that the standard model underestimated customer preferences for features that were not immediately visible.
The researchers say that the findings could have a wide range of applications in industries including consumer products, education, financial services, and housing. Companies could employ the model to predict which product features customers value the most—even when they aren’t fully informed about them.
Without these insights, companies face a universe of potential features to consider. But with them, they can employ just a few promising A/B tests to measure customer responses. As more data become available and usable by businesses and researchers, the method could open up possibilities for more strategic and targeted marketing decisions.
Jason Abaluck, Giovanni Compiani, and Fan Zhang, “A Method to Estimate Discrete Choice Models That Is Robust to Consumer Search,” Journal of Political Economy, July 2026.
Your Privacy
We want to demonstrate our commitment to your privacy. Please review Chicago Booth's privacy notice, which provides information explaining how and why we collect particular information when you visit our website.