Understanding the mechanisms that drive choice is an age-old challenge. Consider the fundamental problem a retailer must solve in stocking a store: a large assortment will likely draw a larger set of customers, but will be substantially more expensive to carry and will also likely introduce cannibalization between products. A smaller assortment might not carry these risks, but may fail to attract a sufficiently large group of customers. One can begin to address this tradeoff by successfully modeling how customer purchase behavior varies as a function of the assortment offered to the customer. Building such models have been as much an art as a science, limiting scope to situations with a small universe of well-understood products. In stark contrast, retailers buy millions of products every season, most of them new to the market.
Celect is able to build predictive models of choice for the gigantic, poorly-understood product universes. We achieve this goal through the development of a new, easily distributed, non-parametric approach to modeling choice that utilizes large quantities of data and distributed computational resources. The key insight to our approach is to treat all transactional data as indicative of an implicit comparison between products. For instance, if you purchased a red jacket when I showed you jackets in red, white, and blue, you’re telling me that you prefer red to blue or white. More generally, such comparisons are ubiquitous in transactional and behavioral data both online and in the store, and we use large volumes of comparisons to build accurate models of choice. Coming full circle, we can now use these models to understand the precise set of products to place at a given store, with a view to maximizing revenue while minimizing assortment breadth, SKU count, and risk to the retailer.
In this talk, we will walk through an innovative new approach to machine learning that seeks to model and learn customer choice patterns and preferences from sparse transactional data. We will then discuss how this approach helps retailers build hyper-local product assortments that are personalized to the foot traffic at each store, while simultaneously reducing assortment complexity and discovering new, surprising opportunities for growth.
Vivek Farias is chief technology officer and co-founder of Celect. He is the Robert N. Noyce Professor of Management at MIT’s Sloan School. His research has led to numerous innovations in operations, supply-chain, and yield management. Prior to academia he worked in algorithmic finance. He received his PhD in electronic engineering at Stanford.
©2015, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • firstname.lastname@example.org
Apache Hadoop, Hadoop, Apache Spark, Spark, and Apache are either registered trademarks or trademarks of the Apache Software Foundation in the United States and/or other countries, and are used with permission. The Apache Software Foundation has no affiliation with and does not endorse, or review the materials provided at this event, which is managed by O'Reilly Media and/or Cloudera.