Brick-and-mortar stores use rows of candy, lip balms, and magazines in the checkout aisles to entice shoppers into making additional purchases. Online shopping sites suggest add-on products for the same reason.
New York University’s Xi Chen, MIT’s Will Ma and David Simchi-Levi, and Chicago Booth’s Linwei Xin have developed an algorithm for these online offers that could help retailers raise the number of such impulse purchases. Their study, which involved a collaboration with Wal-Mart’s online grocery division, indicates that if an algorithm were to consider the retailer’s inventory, it could nearly double add-on sales.
The algorithms behind online add-on offers can be sophisticated, drawing on shoppers’ habits and preferences. But they can also be tricky to tailor, as customers don’t always have to register accounts to purchase products online, and they may shop for a wide variety of items. Moreover, retailers can have trouble figuring out what a shopper actually needs. If a shopper buys five T-shirts, a store might suggest other T-shirts for her to buy, when what she really needs is shorts.
Wal-Mart’s online grocery platform makes add-on suggestions that are based on only the items in the customer’s current shopping cart—and it tries to use those to identify what a shopper might be missing. Say someone has purchased cereal but not milk. The algorithm would offer that customer milk, either at full price or possibly discounted.
But the method has a flaw, the researchers say: it doesn’t take inventory into account. If Wal-Mart is low on milk, or even out of it, the algorithm will still offer the customer milk. If she takes the offer, she depletes an item low in inventory or is frustrated to find Wal-Mart has offered a product that isn’t available. Such an offer risks disappointing the current customer, as well as the next person who comes to the site intending to buy milk.
Thus the researchers’ algorithm introduces the notion of a “protection level in expectation,” which ensures that the number of items offered as add-ons does not exceed a percentage of the total number of expected sales for that product for a given period. If Wal-Mart thinks it will sell 1,000 gallons of milk in a day, for example, the algorithm will only offer milk as an add-on 500 times that day, no matter how many shoppers take the deal.
“It can be important to withhold inventory in order to meet the primary demands of future customers, even with no discounts in prices,” the researchers write.
That one insight, when applied to all the grocery products in a supercenter, can add up to a significant sales boost. Compared with Wal-Mart’s current algorithm, which ignores inventory when suggesting add-on products, the researchers’ algorithm would increase sales by 76 percent, they say.
Xi Chen, Will Ma, David Simchi-Levi, and Linwei Xin, “Dynamic Recommendation at Checkout under Inventory Constraint,” Working paper, November 2016.
The decisions of companies, governments, and educators will help to shape the ultimate outcomes of the A.I. revolution.
A.I. Is Going to Disrupt the Labor Market. It Doesn’t Have to Destroy It.
Don’t underestimate the importance of hiring when pursuing mission and profits.
The Key to Balance? Your Team
Doing good doesn’t have to hurt the bottom line.
Customers Increasingly Expect Your Company to Make a DifferenceYour Privacy
We want to demonstrate our commitment to your privacy. Please review Chicago Booth's privacy notice, which provides information explaining how and why we collect particular information when you visit our website.