Preemptive Shipping: How Gilt Predicts Which Customers Will Buy Products It Has Never Sold Before
The world of fashion is notoriously unpredictable. Customer whims are volatile and different regions have varying tastes. Many products that Gilt offers sell out in a matter of days – sometimes minutes – never to reappear in our inventory again.
We use big data (transactions and clickstream) to deal with a “small data” limitation – no previous history for individual items. Introducing new products is a common problem in personalized recommendations sales forecasting. We’ll discuss selecting and preparing data for such an analysis: cleaning manual-entry texts, accounting for human element. We’ll talk about multi-criterion optimization and show calculating cost of different errors in a multi-way confusion matrix.
In this presentation, Igor Elbert, Gilt’s Principal Data Scientist, will focus on methods used to predict product performance, lessons learned, and unexpected side benefits of this project. Intermediate knowledge of predictive analytics is expected but the presentation can be tailored towards beginners or advanced audience by adding or removing specifics about machine learning methods used.
Mr. Elbert has been dealing with big data for over 20 years. From calculating financial risk for Salomon Brothers to tracking movements of millions of items across the supply chain for major brands, Mr. Elbert pushed innovative data analysis to new frontiers.
As VP of Quantitative Analytics for Barnes & Noble Mr. Elbert used a plethora of data to offer his customers a unique in-store and digital experience.
Having joined Gilt.com as Principal Data Scientist, Mr. Elbert is supporting Gilt’s mission to create the most exciting, curated shopping experience that helps company’s customers find and express their style.