Data Science

The Client

A leading North American video rental company managed 40 thousand automated kiosks throughout the United States and Canada processing over 2 million rentals per day. The company lost over $100 million annually due to fraudulent activity.

The Problem

Identify good vs fraudulent events. With volume of over 2 million rentals per day, identifying good vs fraudulent activity was not possible using their existing systems.

The Solution

We designed and developed two regression algorithms to better identify fraudulent activity in real-time.  The first algorithm focused on the established customer, and the second algorithm focused on the new customers. The output of these two algorithms quantified the likelihood of fraudulent transactions. Transaction data was integrated with multiple other data sources to create accurate quantifications.

Data Collection & Automation

Retail Transactional Data

Demographics data via 3rd party

Retail Store Metadata

Customer Metadata

The Results

Customer achieved annual savings of $25 million by utilizing data science techniques to help identify and prevent fraudulent transactions.

At a glance

INDUSTRY

Retail

CHALLENGE

Need to reduce losses by minimizing fraudulent transactions. 

SERVICES USED

Data mining

Data science

Automated data collection

RESULTS

Implementation of two regression algorithms that determine probability of fraudulent transactions resulted in annual savings of $25 million.