Predicting Service Delivery Times at the Britt Hunt Company

Introduction It was a crisp fall day in 2012 as Doug Sanford sat perplexed in his office in Nashville, Tennessee. As Director of Distribution for The Britt Hunt Company, he had made significant headway in improving the scheduling of customer deliveries. Creating efficient driver schedules was a long-standing Achilles heel for Doug's firm. Efficiency in scheduling was a critical factor in managing delivery costs; a significant portion (54-percent) of the firm's cost structure. From a competitive perspective, logistics had become a significant competitive weapon among food distributors. While Doug's scheduling work had improved performance, one critical problem remained, namely predicting the time it took a driver to unload and stock pizzas (known as "stop time") at the various stores on his or her route. Sipping a cup of coffee, Doug reviewed the situation. What should he decide to do to estimate driver stop times at Britt Hunt? In business since 1992, The Britt Hunt Company employed over 130 employees and had annual sales over $75 million. The firm delivered pizzas products to over 2,600 outlets (convenience stores, gas stations and entertainment venues) across 15 states of the US. Every day company drivers arrived at customer locations, unloaded pizzas products from their trucks, inventoried and stored pizza products, processed paperwork and then drove to the next location. Differences in drivers, store locations, store layout (doors and freezer types) and quantity of product delivered led to variation in the time it took to service a store. This variation led to uneven loading of drivers and the need to "pad" the planned delivery time to avoid overloading drivers. As Doug worked at his PC, he felt stymied. He had installed new route scheduling software in 2011 to improve the scheduling of drivers. Overall, the software was a major step forward and provided significantly improved schedules. The system comprehended drive time between stores. However, Doug found large variations in the time drivers spent at stores and this compromised the effectiveness of his schedules. No obvious answers for the variation stood out as there were a bewildering set of possible explanations. Doug needed to identify predictor variables and a prediction model that he could use to estimate stop times. With accurate estimates of stop times, his scheduling software could do an even more accurate job. Without reliable estimates of how long a driver needed to spend at a given store, it was hard for Doug to evaluate drivers and set equitable route schedules. To address stop times, Doug sought help from a seasoned consultant and professor in Bill Fredenberger. Working together over several months, the two assembled a set of over 92,000 observations of stop times and a set of possible predictor variables. This data covered deliveries over one year and constituted $58 million of sales. Using this data, they sought a way to predict the length of a driver's stop. The Appendix contains a description of the various columns in the data file available to instructors. The Challenge Doug considered a number of analyses that he could pursue. For starters, Doug realized that he could use Excel spreadsheet software for descriptive statistics to see what variables predict stop times. He knew that Excel pivot tables and charts were especially effective in doing this. Next, Doug could perform bivariate statistical methods such as correlation. Clearly, stop time was the dependent variable. He had worked with Bill to identify a set of possible predictors that were available in the data file. From his degree studies, Doug realized that techniques such as correlation, t-tests or ANOVA might be useful in identifying significant predictors. At a deeper level, Bill suggested that Doug could create a predictive model using regression analysis. Both Bill and Doug realized the challenge of identifying which predictors to include. …