When you shop online, it is common for retailers to offer additional items in a bundle to try to increase sales. For instance, if you are buying towels, the seller may offer matching washcloths. Or if you are buying an airline ticket, you may be asked if you also want to purchase inflight wifi and premium seating. If this “bundle” is appealing to you in terms of the items offered and the price, you might be motivated to buy it all. If not, the items or services are left on the table, eventually getting marked down even more.
With the online market projected to grow 57 percent from 2013 to 2018, retailers have the potential to significantly increase their profits through bundling. This strategy can be beneficial for customers too if they are presented with desirable items they otherwise may have missed ― and at better prices. The key is creating an attractive enough bundle to incentivize the buyer to click “add to cart.”
One way to incentivize buyers is to personalize the bundle to their interests and needs. In this digital age, stores have your shopping history, so they know what you’ve previously bought and your shopping journey. If you are a new customer, they can ask you questions about preferences, understand intent from other data sources, see what is trending, and look at what people like you have bought. All of this data can be used to offer the right bundle at the right price to make the sale. While this sounds like an obvious answer, it is a new frontier for retailers.
I have been focusing on this area with IBM researcher Pavithra Harsha, other IBM collaborators, and one of my PhD students, Anna Papush, who also conducted an internship with IBM on this work. Our goal was to construct a new demand model that combines the traditional approach of online shopping with personalization as well as an optimization model for offering each type of shopper (existing customers and new customers) a discounted bundle on relevant items that combines the retailer’s goals of profit maximization and inventory balancing.
We first looked at generic willingness to pay computed using demographics, transactions, and loyalty information. We knew that a one-size-fits-all approach wouldn’t work so we sought to capture the different parts of the population. We factored in shopping journey attributes, context and intent. We then used in-session information to estimate a confidence score. That number, combined with willingness to pay, led to a personalized bundle propensity-to-buy score.
Using optimization and machine learning, we built a model for personalized pricing and product recommendations that leverages knowledge of inventory at risk for markdown. We designed heuristic methodology for adaption to real-time implementation, and we tested the methodology through simulation and work on real online data from a large retailer.
The result was a novel demand modeling and optimization approach that can increase revenue by around 5 percent. This improvement comes from personalized relevant recommendations that are more likely to motivate customers to buy the bundle as opposed to the individual item they were originally seeking. This benefits customers too, as they will find more interesting and relevant items ― that they might not have been aware of ― at better prices, leading to a better shopping experience.
Our model can be used by any retailer from airlines to department stores that want to offer additional items or ancillaries to customers before they check out. We recently filed a joint patent for this work with IBM and MIT.
While our model is groundbreaking, it is just the beginning. We’re working on additional tools to continue improving the online shopping experience for both retailers and consumers.
Georgia Perakis is professor of operations management, operations research and statistics at MIT Sloan School of Management. In recognition of the research collaboration in this area, Perakis received two Faculty Awards from IBM.
