Back to Data Consulting

Taco Bell

Taco Bell is the nation’s leading chain of Mexican-inspired fast food restaurants. As of 2018, Taco Bell served over 2 billion customers each year at roughly 7,000 restaurants, of which 93% were owned and operated by independent franchises and licenses.

Problem

Taco Bell observed high demand from its customers but faced decentralized operations from the many independent restaurant managers. This resulted in the challenge of allotting the right resources to the various locations. As such, Taco Bell engaged with us to create and improve prediction models for the demand of their products so they could organize their supply chain accordingly.

Methodology

Our goal was to predict product demand for over 200 regular, seasonal, and promotional menu items for about 6,000 stores across the nation to plan staffing and food preparation more efficiently. We leveraged Prophet, Facebook’s open source time series forecasting library, to build generalized additive models that consider the overall trend, seasonality, and irregular holiday effects on the demand. We improved these baseline models by incorporating multiple types of seasonal granularity, tuning their Fourier orders, and incorporating exponential moving averages. Finally, we dealt with a lack of data for Limited Time Offer items by classifying them into price buckets with highly similar products and rescaled predictions based on their sales after a week.

Results

We delivered models to predict the sale quantities of each product at a given store over 15 minute time intervals for up to 6 weeks in the future. Our deliverables helped Taco Bell plan for daily staffing, ingredient inventory, and food preparation forecasts at each store. We predicted sales data within 10% of actual sales data and decreased the time-overhead of Taco Bell’s existing model by over 200%.