Grocery stores run price promotions all the time. You see them when a particular brand of spaghetti sauce is $1 off or your favorite coffee is buy one get one free. Promotions are used for a variety of reasons from increasing traffic in stores to boosting sales of a particular brand. They are responsible for a lot of revenue, as a 2009 A.C. Nielsen study found that 42.8% of grocery store sales in the U.S. are made during promotions. This raises an important question: How much money does a retailer leave on the table by using current pricing practices as opposed to a more scientific, data-driven approach in order to determine optimal promotional prices?
The promotion planning tools currently available in the industry are mostly manual and based on “what-if” scenarios. In other words, supermarkets tend to use intuition and habit to decide when, how deep, and how often to promote products. Yet promotion pricing is very complicated. Product managers have to solve problems like whether or not to promote an item in a particular week, whether or not to promote two items together, and how to order upcoming discounts ― not to mention incorporating seasonality issues in their decision-making process.
There are plenty of people in the industry with years of experience who are good at this, but their brains are not computers. They can’t process the massive amounts of data available to determine optimal pricing. As a result, lots of money is left on the table.
To revolutionize the field of promotion pricing, my team of PhD students from the Operations Research Center at MIT, our collaborators from Oracle, and I sought to build a model based on several goals. It had to be simple and realistic. It had to be easy to estimate directly from the data, but also computationally easy and scalable. In addition, it had to lead to interesting and valuable results for retailers in practice.
Partnering with Oracle, we began by mining more than two years of sales and promotions data from several of Oracle’s clients. Using that data, our team developed various new demand models that captured price effects, promotion effects, and general consumer behavior. For example, when paper towels are promoted one week, a consequence is that people stockpile paper towels. Not surprisingly, the effect of a pricing promotion on paper towels the next week is smaller. Our model took that behavior into account. Furthermore, we developed an optimization model that determines the promotion schedule for every item fast.
The first formulation modeled demand “exactly”. Nevertheless, it proved extremely difficult for that model to solve problems in practice. As a result, we created a simpler version that captures 90+% of the complicated version and can solve practical problems. This simpler version can run on accessible software programs like Excel and provides answers in milliseconds. It allows product managers to test various what-if scenarios easily and fast – and be the final decision-makers about promotional pricing.
As for how it works in practice, the simple model is highly effective. When we compared that model with what is currently implemented, we found an average of 3-10% improvement in profits. With typical retail store margins close to 1.9%, promotions can contribute to a significant portion of stores’ profits. For instance, a 5% increase can mean $5 million for retailers with annual profits of $100 million.
So far, together with our Oracle collaborators, we have received very positive feedback on this model and have filed patents for this work. The model has a strong mathematical foundation and can be used by any retailer in any industry. It could be a game changer for retailers, as they seek to optimize promotion pricing.
Georgia Perakis is professor of operations management, operations research and statistics at MIT Sloan School of Management. In recognition of their work in this area, Perakis and her team of students from the Operations Research Center at MIT as well as her collaborators at Oracle received the 2014 Informs Service Science Best Student Paper Award and the 2015 NEDSI Best Application of Theory Paper Award. The team also was also selected as a finalist for the INFORMS Revenue Management & Pricing Section Practice Award in 2015.