Prescriptive Analytics for Large Scale Retail

Project Summary: A platform for prescriptive analytics based on transaction data for large retail stores. The platform provides multiple machine learning features, such as, demand forecasting, hidden demand analysis, clustering, and anomaly detection.

Project Duration: 6 months

Team Size: 5 team members, 2 data scientists

Industry: Retail

Customer

US manufacturing company which is a global leader in retail hardware and software equipment.

Business Case

The company provides a prescriptive analytics platform for their large scale retail customers. The platform aggregates transaction data which is used to identify fraud, meet regulatory compliance, and improve efficiency. It needed to provide a diverse set of machine learning features which can be applied to different retail customers. The data of each customer may differ from the rest, so the platform and its machine learning features needed to be scalable across different customers.

Solution

Our data science team implemented a set of machine learning features for the platform which could be used directly by the users from different retail companies. Some of the ML features included:

  • Demand forecasting and hidden demand analysis for products with intermittent demand. The analysis helped retailers identify days on which they should increase the amounts of stock of particular products in order to maximize their sales.
  • Clustering routines for user defined datasets.
  • Anomaly detection routines for user defined datasets.

The main challenges of implementing the solution were:

  • Scalability of the machine learning features for different customers and use cases.
  • Big and messy unlabeled data that was hard to analyze.

Tools & Technologies

  • Google Cloud Platform
  • Google Cloud BigQuery
  • Python
  • Spark (Google Cloud Dataproc)
  • Google Cloud Composer
  • LightGBM

My Role & Responsibilities

I was the technical lead of the data science team responsible for:

  • Technical consultancy and business analysis
  • Data science team management
  • Data analysis
  • Rapid prototyping of ML features
  • Machine learning research & development

Results

The machine learning features were successfully implemented and integrated into the platform. The platform is currently used by multiple large scale retail customers.