H2O.ai sees machine learning as the answer to firms' growing big data collections

Tools

Financial firms' growing collection of structured and unstructured data has the potential to yield new visibility into business processes. But artificial intelligence proponents say the finance industry will need to embrace machine learning to convert vast amounts of big data into timely and actionable insight.

Research firm International Data Corp. (IDC) has forecast that the market for big data will grow at an annual rate of 23 percent through 2019. In financial services, that growth is even faster. Securities and investment services tied with banking as the industries with the fastest growth rate in big data, with compound annual growth rates of 26 percent.

"The biggest challenge is inertia. Companies are used to the old rules-based models they had in the past," says Vinod Iyengar, director of marketing for H2O.ai.

Machine learning solutions essentially scan vast data warehouses to identify the data points that are indicators of the type of results a firm wants to identify. They then apply algorithms to identify the patterns between relevant data points.

"Machine learning does the same thing a human can do, but it can look through multiple interactions between hundreds or even thousands of data points to come up with a pattern that says if these X things happen together, we come up with an outcome which is Y," Iyengar said.

In financial services, Iyengar says H2O has seen clients begin to use machine learning tools to create alternative FICO scores for assessing creditworthiness, for fraud detection, to predict customer churn and more.

A key advantage to using machine learning with big data is that it often replaces solutions that were built based on a small sample of data, and consequently not as accurate. Machine learning can evaluate a wide spectrum of data, and then cut away data that is not relevant to the models a company is trying to build.

Machine learning, by nature, also offers advantages in speed, Iyengar says. Models can be built quickly, and easily modified when a new anomaly appears that doesn't fit pre-existing models. For example, in fraud detection, as fraudsters develop new techniques for attacking a company's network, the characteristics of the new type of fraud can be quickly identified and algorithmic models can be quickly updated to take into account the new data points.

While not in financial services, H2O client Cisco builds or updates 60,000 models a week, Iyengar says, because of the ease of creating and modifying models. The ability to modify models quickly and easily also has applications for heavily regulated industries, like financial services, where shifts in regulations can be easily accommodated, Iyengar noted.

"We see a lot of use cases come in from companies that often don't see machine learning as a solution," Iyengar said. "If they have the data they can predict any kind of outcome."

Related Articles:
Genpact taps RAGE to integrate artificial intelligence into risk and middle- and back-office functions
Publicis.Sapient partners with Lucid to apply causal artificial intelligence to finance