Optimizing Product Rankings for Competing Business Objectives Optimizing Product Rankings for Competing Business Objectives
Editor’s note: Ali Vanderveld is a speaker for ODSC West 2022 coming this November 1st-3rd. Be sure to check out her... Optimizing Product Rankings for Competing Business Objectives

Editor’s note: Ali Vanderveld is a speaker for ODSC West 2022 coming this November 1st-3rd. Be sure to check out her talk, “Optimizing Recommendations for Competing Business Objectives,” there!

When customers search or browse for products at Wayfair, their experiences are shaped by algorithms developed by scientists at the company to determine the optimal selection to surface at every stage of their shopping journey. Recommender systems traditionally tackled this problem by optimizing metrics related to near-term customer satisfaction, including clicks to a product detail page, add to cart actions, or attributable purchases.

However, an e-commerce marketplace is a complex ecosystem with multiple constituents: these can include customers, suppliers, and the e-commerce company itself. Each of these constituents has their own objectives that may or may not be aligned with each other. For example, a customer might want to purchase a particular sofa, but that item may not be stocked in the closest fulfillment center, while there are many very similar sofas available nearby. To reconcile competing objectives, recommender systems ideally should be able to do more than optimize metrics related to short-term clicks and orders – they should also make “business aware” recommendations that take into account considerations related to long-term customer satisfaction and the overall health of the business. “Business aware” metrics can encompass many facets: the e-commerce company’s revenue and profitability, the health of a company’s suppliers, the number of goods that show up damaged at customer residences, and the carbon footprint associated with shipping. Recommender systems can factor in all of these objectives while surfacing relevant product recommendations to customers.

In an ideal world, recommender systems would optimize for a state of optimal “Pareto efficiency” – a state where no single improvement in any one business objective could be achieved without some other objective being rendered worse off. However, scientists at Wayfair and other e-commerce companies have to take into account other important practical considerations: these include factors like latency, resources needed to maintain the system, and whether it is scalable across an organization. Ideally, we would be able to train one model that optimizes for competing objectives while arriving at the frontier of all Pareto-efficient solutions. However, we would still ultimately have to determine where we want to live along the Pareto frontier. For example, how much of a short-term conversion rate reduction are we willing to have in order to maximize our long-term profit?

Over the last two decades, academics and experts in the industry have utilized a variety of approaches to solve this problem. Each of these approaches has their benefits and drawbacks, and none of them can serve as an “off-the-shelf“ solution for any company’s peculiar needs. Moreover, in order to measure the effectiveness of such strategies in online testing, one needs to move away from solely relying on short-term metrics related to customer satisfaction, to consider metrics of long-term customer retention. Instead of just asking the question, “Will this customer buy a particular sofa during a browsing session?,” we are also beginning to ask, “What can we do to make this customer keep coming back to Wayfair and buying products in the long term?”

During my talk at ODSC West 2022 titled “Optimizing Recommendations for Competing Business Objectives,” I will provide an overview of the problem of training “business-aware” recommender systems, what makes it difficult, how we have been addressing it at Wayfair, and the lessons that we have learned so far.

About the author/ODSC West 2022 speaker:

Ali Vanderveld is a Senior Staff Data Scientist at Wayfair, where she serves as a technical leader for machine learning, currently leading the development of novel search and recommendation technologies. Prior to Wayfair, she led a team focused on language AI at Amazon Web Services and was the Director of Data Science at ShopRunner. She has also worked at Civis Analytics, at Groupon, and as a technical mentor for the Data Science for Social Good Fellowship. Ali has a PhD in theoretical astrophysics from Cornell University and got her start working as an academic researcher at Caltech, the NASA Jet Propulsion Laboratory, and the University of Chicago, working on the development teams for several space telescope missions, including ESA’s Euclid.

ODSC Community

The Open Data Science community is passionate and diverse, and we always welcome contributions from data science professionals! All of the articles under this profile are from our community, with individual authors mentioned in the text itself.