Photo by Susan Young

Photo by Susan Young

How did you come to focus on algorithmic pricing?

In my doctoral work at MIT, I was studying optimization, probability, and machine learning, which are essentially mathematical tools that enable us to use data to make better decisions. From there, I realized I wanted to focus on a business application in an industry with the most room for improvement. Online retail, particularly online fashion, definitely fit the bill.

Why is fashion a good place to focus?

The industry faces several challenges that many other retailers don’t. There is quite a bit of heterogeneity in customer style preference. Many products are sold for just one selling season. It’s extremely difficult to predict demand—sometimes demand actually increases as the price goes up due to perceived quality and popularity. These factors make selling fashion products quite different than selling, say, milk or toilet paper. In addition, online retailers have a vast amount of data at their fingertips; solving complex problems with data seemed like a natural fit.

Tell me about the research you did with Rue La La.

Rue La La is in the online flash sales industry, offering designer apparel and accessories at a deep discount for a limited time. Many of its products sell out the first time they’re offered, and a majority of the revenues come from first sales. So the question of how to price new products is an important one, but there’s no historical sales information available to go by.

To address that need, my coauthors and I identified features from past products—such as price and the number of similar products sold at the same time—that we could map to features of new products as a way to predict demand. We also took into account the need to price multiple products in the same category at the same time, since demand for one item is influenced by the price of all the others. After we built and implemented a software tool to make demand predictions and optimize prices, we ran a field experiment to test the impact of our tool and saw a 10 percent revenue increase in the test group. This shows that online retailers can enjoy a substantial gain with a relatively low investment when they go beyond the common approaches of determining price based on intuition, competition, or a standard cost-plus-markup formula.

How does this connect with your research on dynamic pricing?

Many retailers mark down prices at the end of the season as a way to clear inventory. The problem with that is you might have strategic consumers who delay their purchase of a pair of sandals, for example, to take advantage of the discounted price. Whereas if you didn’t use a markdown, then—who knows?—maybe those consumers would have purchased the item at full price.

We suggest that online retailers change their price more frequently at the beginning of the selling season to gauge demand at different price points. If you can go through that process quickly—which is possible in the world of online retail—you can set an optimum price for the rest of the selling season. There’s a fundamental tradeoff: Should I learn something about demand by offering a price that might be suboptimal, or should I offer the price that I think is optimal now to maximize revenue? Our research shows retailers a way to balance this tradeoff between “learning and earning.”

Post a Comment