Turning Up the Heat on Sales Forecast Accuracy
November 19, 2015
Editor’s Note: Every year, 40 or so students in the MIT Center for Transportation & Logistics’ (MIT CTL) Master of Supply Chain Management (SCM) program complete one-year thesis research projects. The students are early-career business professionals from multiple countries with 2 to 10 years of experience in the industry. The research projects are sponsored by and carried out in collaboration with multinational corporations. Joint teams of company people, MIT SCM students, and MIT CTL faculty work on real-world problems chosen by sponsor companies. In this monthly series, we summarize a selection of the latest SCM research. The researchers for the project described below, Vinod Bulusu and Haekyun Kim, analyzed the accuracy of sales forecasts for auto batteries for their MIT Supply Chain Management Program master’s thesis. The sponsor was a manufacturer in the after-market replacement battery business, and the project was supervised by Dr. Roberto Perez-Franco, Research Associate, MIT CTL. For more information on the program, visit http://scm.mit.edu/program.
After being caught off guard by a phenomenal increase in sales in 2013, a company in the auto battery replacement market wanted to improve the accuracy of its sales forecasts. The problem was exacerbated by a relatively long ramp up to new production.
Consumers go shopping for new car batteries when their existing units fail, so helping the company to anticipate failures helps the organization to improve the accuracy of its sales forecasting.
An important determinant of the lifetime of an auto battery is external temperature; for example, it’s harder to start cars in a winter chill and this season generates an uptick in demand. But how do variations in temperature impact sales volumes? The thesis project aimed to help the company answer that question.
The researchers collected point of sale (POS) information from battery retailers, including details such as the vendor, date of sale, zip code, SKU and units sold. Data on maximum and minimum temperatures were also collected, since battery failures tend to occur at temperature extremes. Daily and weekly temp changes were included to correlate with the POS data, as well as variations by metropolitan region.
A SKU was selected as the focus of the research. Important considerations were to make sure that the SKU-related dataset was large enough to assure model reliability, and that it was sold over a wide geographic area in the US that covered a broad range of temperatures.
Based on this data, the researchers built a number of models and selected the one that best reflected temperature variations as a predictor of sales.
Two additional models were created to validate the main model: one that considered three cities, Chicago, Houston, and Los Angeles, and one that encompassed all the cities in the study.
Improvements with age
The model did indeed establish a correlation between sales and temperature. It was found that minimum temperatures represent the best predictor. The model is more robust across time than geography. Also, although the model can predict the direction of magnitude of change in sales volumes, it can’t determine the absolute value of sales.
Based on the results, the sponsor company can refine the model by adding sales and temperature information from various geographies. Battery age is another factor that can be built into the model to make it more accurate. It is possible to calculate this value by carrying out customer surveys in representative metropolitan areas.
Subscribe to Supply Chain Management Review magazine
cutting supply chain costs and case studies in supply chain best practices. Start Your Subscription Today!