DataRobot > Case Studies > How Consensus, a Target subsidiary, simplified data wrangling for machine learning

How Consensus, a Target subsidiary, simplified data wrangling for machine learning

Customer Company Size

Large Corporate

Region

America

Country

United States

Product

Trifacta Wrangler Pro
DataRobot

Tech Stack

Amazon S3
Amazon Redshift
Amazon EMR
AWS VPC

Implementation Scale

Enterprise-wide Deployment

Impact Metrics

Cost Savings
Customer Satisfaction
Productivity Improvements

Technology Category

Analytics & Modeling - Machine Learning
Analytics & Modeling - Data-as-a-Service
Infrastructure as a Service (IaaS) - Cloud Computing

Applicable Industries

Retail

Applicable Functions

Business Operation
Sales & Marketing

Use Cases

Fraud Detection

Services

Data Science Services
System Integration

About The Customer

Consensus Corporation is a subsidiary of Target that simplifies the complex process of selling connected devices. It connects retailers with manufacturers and network operators using a single platform to enable what it calls multiplex selling: bundling technology and service purchases together, such as a smartphone and a data plan, or a smart television and a subscription to a video streaming service. Consensus enables retail stores nationwide to sell these types of connected devices and services through one unified online platform. However, a major risk for retailers selling expensive devices and services is fraudulent customer activity. To address this risk, Consensus adopted fraud prevention as one of its core services.

The Challenge

Consensus Corporation, a subsidiary of Target, simplifies the complex process of selling connected devices. However, a major risk for retailers selling expensive devices and services is fraudulent customer activity. To address this risk, Consensus adopted fraud prevention as one of its core services. Through its automated machine learning-powered online engine, Consensus can alert its retailer clients to high-risk consumers before they purchase expensive devices. To identify potential fraud, Consensus built an advanced data model that leverages huge volumes of disparate data and undergoes routine updates. In order to be able to constantly refine its predictive models and alert their retailer clients faster to potential fraud, Consensus sought out technologies that would allow it to prepare this data faster for use in its machine learning models. The painstaking process of re-engineering SQL scripts took Consensus up to six weeks (on average) to update its fraud detection machine learning model. In addition, the data preparation process required sophisticated knowledge of data science techniques, leaving the company’s product and business intelligence teams unable to perform data preparation tasks on their own.

The Solution

Consensus identified the need for better data preparation and rapid model prototyping and evaluation that did not require advanced data science knowledge. At first, the company attempted to work with several third-party data preparation solutions, but found them either expensive, unwieldy, or complicated to deploy. Finally, Harrison Lynch, Senior Director of Product Development for Consensus, discovered the DataRobot automated machine learning platform and free Trifacta Wrangler solution. Trifacta’s user-friendly UI and DataRobot’s ability to rapidly build and deploy machine learning models helped differentiate both technologies from the competition. Consensus then selected DataRobot and Trifacta Wrangler Pro on AWS to get more power and more connectivity to wrangle even more data and quickly create machine learning models. Trifacta Wrangler Pro & DataRobot are AWS–deployed solutions that seamlessly access data stored on AWS, including Amazon S3 and Amazon Redshift. In addition, Trifacta Wrangler Pro leverages Amazon EMR (Elastic MapReduce) to process the data.

Operational Impact

Consensus was able to use Trifacta to wrangle large amounts of structured historical data stored in Amazon S3 and more accurately deliver machine learning models in less time with DataRobot compared to traditional methods.
Trifacta helped solve the problem of uploading the most accurate data into Consensus’s fraud detection models quickly, without the cost and potential inaccuracies associated with relying on manual data preparation or traditional languages such as SQL, R, or Python.
DataRobot helped solve the problem of high numbers of false positive predictions that were hurting customer experience at the point of sale as well as detecting potential sources of fraud with higher accuracy.

Quantitative Benefit

24% gain in True Positive detection
55% decrease in False Positives
19% gain in overall financial performance
Speed of Deployment from three to four weeks of combined effort to less than eight hours to implement new models and respond to business changes
Accelerate Model Training from two to three days and at least two people to prepare data for one model, to training and validation in only two to three hours by one person