Capital One study cites ML anomaly detection as top use case
The bank itself is making ML a key part of its digital transformation strategy, deploying the technology to locate aberrations, detect fraud and support marketing.
A Capital One-commissioned machine learning study identified anomaly detection as the top priority for data managers in the next one to three years.
The report, published last month, found 40% of the 150 data management decision-makers polled said they planned to deploy ML to automate anomaly data detection, making that the top application for the organizations' evolving data and analytics strategies.
ML anomaly detection edged out automatic application and infrastructure updates and meeting regulatory and privacy requirements for responsible AI use as top data and analytics concerns. Forrester conducted the survey on Capital One's behalf.
ML applications at Capital One
Capital One has stolen a march when it comes to adopting ML for anomaly detection. Amid the current phase of economic uncertainty, the financial services firm uses ML to look for signs of trouble across its businesses, which include credit cards, auto loans and business lines of credit.
During the company's Oct. 27 third-quarter earnings call, Rich Fairbank, CEO at Capital One, cited ML's role in looking for aberrations and anomalies.
"Real-time machine learning-driven monitoring is really important in any environment. But it's really important in an environment like this," Fairbank said. "We expect anomalies to happen, [and] we expect some of our segments to be stressed. So we're on the lookout, and we have machine learning to help us."
Most companies have two or fewer years of ML experience, but data managers plan to prioritize anomaly detection.
ML also supports credit card fraud defense. Capital One uses home-grown and open source ML algorithms to detect anomalies and automatically create defenses, said Dave Kang, senior vice president and head of data insights at Capital One.
The technology will see an expanding role in cybersecurity, where technical personnel deal with alert fatigue, said Jay Pasteris, CISO and CIO at GreenPages, a cloud and cybersecurity services company based in Kittery, Maine.
ML continuously learns as it encounters trillions of events, which helps it distinguish signal from noise. ML also helps organizations react to alerts in real time and provides better observability, he added.
"I think ML and AI are going to grow and play a big part in the securing of organizations," Pasteris said.
As for other applications, Capital One uses ML to bolster demand generation. Fairbank cited "mass-customized, machine learning-driven marketing" as part of the reason the company has been able to invest more in cultivating new customers. That marketing push aims to expand the company's heavy-spender credit card business and digital national bank, for example.
ML strategy can take years to mature
Taking ML from the experimentation stage to a core technology doesn't happen without effort. Forrester Research's ML study found that 77% of organizations had two years or less experience in developing and releasing ML applications, noting that a mature ML strategy can take three or more years to crystalize. Obstacles include silos between data scientists and practitioners, which 57% of the survey's respondents cited as slowing ML deployments. Meanwhile, 38% of the decision-makers polled said data silos across the organization and external data partners also hinder ML maturity.
ML is increasingly a foundational capability that, much like technology deployment overall, transcends economic cycles.
Dave KangSenior vice president and head of data insights at Capital One
For its part, Capital One employs standardization to break down silos. That means standardizing tools, process and platforms so data scientists and engineers can more easily identify and access data, Kang said. They can then "build on the foundations we've established to deploy ML models" he noted.
Capital One has focused on getting teams collaborating on the same technology stack and prioritizing reusable components and frameworks across the company's ML efforts. Kang said his team created an ML platform that provides "governed access" to those components along with algorithms and infrastructure. This approach lets non-data science/ML practitioners use ML for business decision-making.
"As we continue to scale machine learning across the enterprise, we're leveraging best practices and approaches like standard platforms, democratization, libraries for features and algorithms, and continuous learning and training," Kang said.
ML plays part in broader transformation
The company's ML efforts appear set to continue despite the macroeconomic trends.
"From Capital One's perspective, ML is increasingly a foundational capability that, much like technology deployment overall, transcends economic cycles," Kang said.
The company's ML investment is part of a broader, long-term technology transformation journey, he noted. Overall, the company invested $349 million in communications and data processing technology in Q3 -- a 9% year-over-year increase.
"Technology investment helps our revenue growth and helps drive productivity improvements," Fairbank said. "And beneath the surface of the high-level of investment has been significant productivity gains from modernizing our tech stack, eliminating legacy vendor costs, driving customers to digital and driving more automation in the company."
Dig Deeper on IT applications, infrastructure and operations
