Every year, there are a few items of clothing that become hot. For example, last fall, a Zara coat seemed to become a “must have” item. The coat even had its own Instagram page with more than 8,000 followers. Many factors contribute to this phenomenon like celebrities ― and people with large social media followings ― wearing the “hot” item.
When we have detailed social media data, it is relatively easy to identify patterns of influence to predict these trends. But what happens when we don’t have social media data? After all, social media platforms charge tremendous fees for access to that information. Can we use traditional data to detect underlying trends between groups of consumers and improve demand estimation? If so, can we use that information to optimize personalized promotions to increase profits, and also to present “the right individual with the right item at the right price?” Personal Queue
In a recent study, I looked at these questions with MIT Operations Research Center PhD students Lennart Baardman and Tamar Cohen and collaborators from Oracle Retail. We found that the answer to both questions is: yes. We began our study by building a customer demand model and algorithm that incorporates customer-to-customer trends or influences. We then applied the information about customer demand to make promotion decisions. With this method, profits increased between 5-12%. The model can be used by any retailer of any size for any product.
The model works by inputting traditional data like store locations, the types of people who shopped at these locations, the timing of when a specific product was purchased, and how much each customer spent. The goal was to find a way to target the right customers with promotions so that they purchase and influence others to purchase, maximizing the profit.
We found that customer-to-customer trends can be dependent on item, time, and promotion. We also found interesting patterns related to geography. While it might seem logical that consumers in big urban areas would influence consumers in smaller areas, we discovered that this isn’t always the case. When we looked at Ohio, it turned out that Columbus influences the Dayton South area, which then influences Cleveland. Columbus is a big city, but it was surprising to see that the smaller Dayton South influences Cleveland.
Based on that data, it would be better to target Columbus first, rather than Cleveland or Dayton South, because it would be like hitting two birds with one stone. If you can motivate a group of key influencers in Columbus, you can get others to buy the same products in Dayton South and Cleveland and beyond. With this strategy, you get more bang for your marketing buck.
The takeaway is that targeting the largest influencer is not always optimal. Sometimes it is better to target an influencer to the largest influencer.
We also found that it’s important to target the right customers with promotions so that they purchase and influence others to purchase, maximizing the profit. This means doing promotions only at specific time periods, limiting the total number of promotions, and using a different limit on the number of promotions at each period.
It is clear that, even without social media data, we can still understand the influence of certain individuals and groups. Even better, we can use that information to target and motivate others to improve overall profitability.
Prof. Georgia Perakis is the William F. Pounds Professor of Management and a Professor of Operations Research and Operations Management at the MIT Sloan School of Management. She also is the faculty director of the MIT Executive MBA Program.