Washington University researchers provide boost to online sales

Online retailers now have a powerful new tool on their virtual shelves that could help them significantly increase revenue and profitability.

Shoppers in the U.S. spent $513.61 billion online in 2018, up 14.2 percent from 2017, according to U.S. Department of Commerce estimates. 

Ecommerce sites such as Amazon, eBay, Walmart and others already have sections such as “Customers also shopped for” or “Compare with similar items” to show shoppers alternative products. But what if a better system could be developed that populates those sections with products that would generate a lot more money for the sellers?

Development of a better product recommendation system — one built to maximize revenue per customer that captures customer purchasing behavior and taps into historical sales data — was behind “Taking Assortment Optimization from Theory to Practice: Evidence from Large Field Experiments on Alibaba,” a research paper by Jake Feldman and Dennis Zhang of the Olin Business School.

Feldman and Zhang conducted an experiment involving more than 5 million online shoppers. The customers were associated with Chinese online retail giant Alibaba, which sold more than $485 billion worth of merchandise in 2016, surpassing Walmart as the world’s largest retailer.

Dennis Zhang, assistant professor of operations and manufacturing management at Washington University’s Olin School of Business

Feldman and Zhang tested two approaches to display alternative products. The first approach tested was Alibaba’s machine learning model, in which the products displayed were based on computer estimates of customer buying patterns. The second approach was based on Feldman and Zhang’s new model, whereby the products displayed were based on customers’ buying patterns as determined by their purchasing history. Both approaches were tested on two separate Alibaba sites that connect third-party retailers to customers.

What Feldman and Zhang found was that their approach generated 28 percent higher revenue per customer visit compared to Alibaba’s machine learning approach that used similar features.

The superior performance of the system can be attributed mainly to the fact that it displays more expensive products to customers, but also products that customers are more inclined to buy based on their previous purchases.

“We are hopeful that our work inspires other companies to consider a choice-model-based approach within their product recommendation system and that it also encourages other researchers in operations management to seek out avenues to implement their algorithms in practice,” Feldman and Zhang said.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.