Forcasting customer demand at Taco Bell

Like most quick service restaurants, Taco Bell understands the quantitative trade-off between labor and speed of service. More than 50% of the $5 billion company's daily sales come from the 3-hour lunch period. Customers don't like to wait more than 3 minutes for services, so it is critical that proper staffing is in place at all times.
Taco Bell tested a series of forecasting models to predict demand in specific 15-minute intervals during each day of the week. The company's goal was to find the technique that mini­mized the average square deviation between actual and pre­dicted data. Because company computers only stored 6 weeks of transaction data, exponential smoothing was not consid­ered. Results indicated that a 6-week moving average was best.
Building this forecasting methodology into each of Taco Bell's 6,500 stores' computers, the model makes weekly pro­jections of customer transactions. These in turn are used by store managers to schedule staff, who begin in 15-minute increments, not one-hour blocks as in other industries. The forecasting model has been so successful that Taco Bell has documented more than $50 million in labor cost savings, while increasing customer service, in its first four years of use.