Business Impacts
25%
Reduction in food wastage
$4.5
Million savings per month
25%
Improvement in model accuracy
Customer Key Facts
- Location : North America
- Industry : Restaurants
Problem Context
The customer’s legacy statistical-based modeling approach forecasts demand in terms of dollars earned, and item quantities sold at a daily and sub-hourly level. The forecast has to be made several weeks in advance to cater to the business requirements. Legacy methods, however, struggled to perform in a dynamic and fast changing environment.
Challenges
- Sub hour level forecasting
- Scaling to 2,000 stores and more than 400 products
- Seasonal variations in data
- Scarce and missing data points
Technologies Used
AWS Lambda
Amazon S3
Amazon Athena
Amazon EMR
Amazon Aurora
Amazon Sagemaker
TensorFlow
Jenkins
Machine Learning Forecasting Model for a Quick Service Restaurant
The customer wanted a state of the art demand forecasting solution with a dashboard to receive updates on a periodic basis to have better control on demand fluctuations in order to have an efficient supply chain, which in turn would reduce costs, food wastage, and optimize available warehouse space.
Solution
Quantiphi built a forecasting engine to predict sales at a daily and intra day basis for multiple different time horizons. Our flexible models helped the client with efficient warehouse management and manpower scheduling. The solution also improved the estimates of the suggested quantities of ingredients to be ordered from the distribution centers. The solution was scaled to more than 2000 stores.
Result
- Quantiphi’s demand forecasting models were able to perform better than the customer’s legacy system by about 25 percent in terms of error reduction
- Sensitivity analysis capability shown by the models based on the previous occurrences of the external factors increase sales and improve revenues
