H2O.ai Case Studies Capital One Uses H2O for Mobile Transaction Forecasting and Anomaly Detection

Edit This Case Study Record

	Capital One Uses H2O for Mobile Transaction Forecasting and Anomaly Detection H2O.ai

Capital One Uses H2O for Mobile Transaction Forecasting and Anomaly Detection

H2O.ai

Technology Category	Analytics & Modeling - Predictive Analytics Analytics & Modeling - Real Time Analytics
Applicable Industries	Finance & Insurance
Applicable Functions	Business Operation
Use Cases	Predictive Maintenance Real-Time Location System (RTLS)
Services	Data Science Services
Challenge	Capital One's mobile app is a popular platform for customers to perform transactions, with up to 5,000 customers logging in every minute. This high volume of usage means that even small outages need to be identified and resolved quickly to prevent service disruptions. The bank's technology operations group monitors all critical systems and platforms and sets up alerts based on company policies. However, setting up alerts for volume anomalies, such as a drop or spike in transaction volume, proved to be a challenge. Traditional methods of calculating volume anomalies were not scalable and required a lot of coding, development, and oversight to manage. Read More
About Customer	Capital One is a well-known American bank that values the power of information and technology to deliver highly customized financial products to consumers and business customers. The bank is recognized for its innovative approach, particularly in making banking secure, convenient, and user-friendly. Capital One's mobile app is rapidly becoming the preferred channel for customers to perform transactions, with up to 5,000 customers logging into the platform every minute. The bank has a dedicated technology operations group that monitors all of the bank's critical systems and platforms to ensure the mobile app is up and running at all times. Read More
Solution	To address the challenge of detecting and alerting on volume anomalies, Capital One's data scientists turned to platform engineering and open source technology. They used Sparkling Water, a tool that combines the fast, scalable machine learning algorithms of H2O with the capabilities of Apache Spark, for rapid prototyping and ad-hoc experimentation. H2O's advanced capabilities for in-memory processing matched Capital One's big data environment needs, and its support for Python, Spark, and Scala enabled a unified coding pipeline for the bank's data experts. The team started with the Generalized Linear Model (GLM), but found that the Gradient Boosting Machine (GBM) provided greater flexibility. GBM allowed the team to account for trends and seasonality in their mobile application usage, and to filter and exclude data as needed. To productize the solution, the team built a scalable and repeatable pipeline using cloud-based, open source platforms and tools. Read More Log in to view content
Contents

Technology Category

Analytics & Modeling - Predictive Analytics

Analytics & Modeling - Real Time Analytics

Applicable Industries

Finance & Insurance

Applicable Functions

Business Operation

Use Cases

Predictive Maintenance

Real-Time Location System (RTLS)

Services

Data Science Services

Challenge

Capital One's mobile app is a popular platform for customers to perform transactions, with up to 5,000 customers logging in every minute. This high volume of usage means that even small outages need to be identified and resolved quickly to prevent service disruptions. The bank's technology operations group monitors all critical systems and platforms and sets up alerts based on company policies. However, setting up alerts for volume anomalies, such as a drop or spike in transaction volume, proved to be a challenge. Traditional methods of calculating volume anomalies were not scalable and required a lot of coding, development, and oversight to manage.

About Customer

Capital One is a well-known American bank that values the power of information and technology to deliver highly customized financial products to consumers and business customers. The bank is recognized for its innovative approach, particularly in making banking secure, convenient, and user-friendly. Capital One's mobile app is rapidly becoming the preferred channel for customers to perform transactions, with up to 5,000 customers logging into the platform every minute. The bank has a dedicated technology operations group that monitors all of the bank's critical systems and platforms to ensure the mobile app is up and running at all times.

Solution

To address the challenge of detecting and alerting on volume anomalies, Capital One's data scientists turned to platform engineering and open source technology. They used Sparkling Water, a tool that combines the fast, scalable machine learning algorithms of H2O with the capabilities of Apache Spark, for rapid prototyping and ad-hoc experimentation. H2O's advanced capabilities for in-memory processing matched Capital One's big data environment needs, and its support for Python, Spark, and Scala enabled a unified coding pipeline for the bank's data experts. The team started with the Generalized Linear Model (GLM), but found that the Gradient Boosting Machine (GBM) provided greater flexibility. GBM allowed the team to account for trends and seasonality in their mobile application usage, and to filter and exclude data as needed. To productize the solution, the team built a scalable and repeatable pipeline using cloud-based, open source platforms and tools.

Impact #1	The data team delivers value by providing the monitoring teams with a visual representation of volumes and anomalies. This allows anyone who is triaging an issue or investigating a root cause of a problem to see where the current volume is relative to what’s normal.
Impact #2	The anomaly band, marked with a timestamp, helps a monitoring engineer find the exact time when an anomaly has occurred, allowing them to measure and correlate it to the start time of incidents and other alerts.
Impact #3	Messaging alerts help warn the operations team when a problem is detected.

Impact #1

The data team delivers value by providing the monitoring teams with a visual representation of volumes and anomalies. This allows anyone who is triaging an issue or investigating a root cause of a problem to see where the current volume is relative to what’s normal.

Impact #2

The anomaly band, marked with a timestamp, helps a monitoring engineer find the exact time when an anomaly has occurred, allowing them to measure and correlate it to the start time of incidents and other alerts.

Impact #3

Messaging alerts help warn the operations team when a problem is detected.

Benefit #1	In one instance, an anomaly alert was sent when a spike in volume was detected at 11:15pm – with over 20 thousand more users logging in as would typically be expected for that time of night. Incident response teams were alerted promptly at 11:17pm, four minutes before any other incident alarms were triggered.

Benefit #1

In one instance, an anomaly alert was sent when a spike in volume was detected at 11:15pm – with over 20 thousand more users logging in as would typically be expected for that time of night. Incident response teams were alerted promptly at 11:17pm, four minutes before any other incident alarms were triggered.

Download PDF Version

Overview

Capital One Uses H2O for Mobile Transaction Forecasting and Anomaly Detection

Operational Impact

Quantitative Benefit