Enterprise Forecasting Engine
Retail & CPGBusiness Impacts
Potential savings ~ $5MN per month
25% reduction in food wastage
Right sizing inventory orders for restaurants with ~ 1M predictions daily
Customer Key Facts
- Location : USA
- Industry : Retail/CPG
Problem context
The client is among the top 10 fast-food restaurant chains in the USA and operates more than 2,774 restaurants across 47 states.
Their legacy forecasting engine didn’t provide complete visibility into the demand patterns and forecasts generated. They wanted to improve inventory management using the AWS platform and EMR for making proactive decisions.
Challenges
- Finding an exhaustible set of features which affect demand trends for building the model like holidays, weather conditions etc.
- Developing a flexible, parameterized process with no single point of failure which makes it easy to add functionalities in the future.
- Large scale data transformations with time and cost constraints for over 2000 locations and 500 SKUs .
Technologies Used
Amazon EMR
Amazon Sagemaker
Amazon S3
AWS Lambda
Amazon EC2
Amazon Athena
Solution
Quantiphi developed an Enterprise level forecasting engine that generated forecasts for dollar sales, number of transactions, products, and ingredients at daily and 15 mins level with 95% model accuracy.
The solution provided 1M predictions on a 2000 node cluster on a daily basis. Starting with forecasting for one store, it has been scaled to 2000+ branches of the restaurant chain in a fully automated and optimized fashion on AWS infrastructure.
Quantiphi leveraged EMR to load data from the data lake in S3 to the forecast account. We also used transient clusters to roll up the transactional data into daily and 15 mins level using Spark SQL.
The EMR processing runs on a daily job and takes 3-4 hours to complete.
Results
Successfully developed a forecasting engine to forecast demand several weeks in advance to cater to various business requirements by leveraging Amazon EMR.
