NAFIS 2017: Firms Dip Their Toes into Data Lakes for Analytics

Data warehouses in firms today have traditionally grown within business silos, or may have come from acquisitions, said Thomas Angelo, digital collaboration services director at BNY Mellon, speaking on a panel at the North American Financial Information Summit held in New York on May 24. “The problem is, they’re very difficult to scale. If you want to add a new business to a warehouse, it’s difficult to add [because that warehouse is typically] a best-of-breed developed for one solution. But to get that enterprise view across a firm to better identify business opportunities… you need to unify [those warehouses]—and that’s where data lakes come in. They are a unified way for applications to access data across that lake,” meaning that anyone within an organization who is interested in that data can use it to run analytics that inform their decision-making process.

This means that the data lake architecture is often attractive to high-level business executives at firms “because you just dump all your data in raw format into one place” that doesn’t require the same amount of upfront work as data warehouses, said Jean-Pascal Chauvet, chief technology officer for Deutsche Bank’s equities business. However, a lot of work is needed to make use of the data stored in the lake, requiring data scientists who understand a firm’s business, he said.

Another disadvantage is that data tipped into the lake is stored there without context. “You don’t have the lineage of the data, why it exists, and how it came to be,” Chauvet said, adding that there are also access control issues. “If you dump all your data in one place, how do you control the access, if you don’t know the context around the data? In this way, data lakes can very quickly become data swamps.”

Wells Fargo thinks of data lakes as “the ‘place’ where our businesses and those who interrogate the data in the lines of business will now have the opportunity to be able to go to this unified ‘ecosystem’ that pulls together their business-respective data in a common place, in a common way, governed and structured…. It’s not an easy thing to pull together,” said Stephen Harris, senior vice president, head of enterprise data strategy at the bank.

But Harris also warned that a data lake is not a quick, easy solution to all of a firm’s data challenges. “A data lake is not a silver bullet…don’t be fooled by the concept that the data lake is going to be the be-all answer. It’s another approach you can take to servicing the insight needs of your respective business partners,” requiring “multiple drawers” to manage structured and unstructured data. “The lake is an approach, a framework; it will evolve to become a technology ecosystem.”

While applications for data lakes are still evolving, panelists agreed that one successful use case for data lakes is internal and external fraud detection. “You need to be able to marry trading data with entitlement data—who has access to what— with records such as security pass swipes. On top of that you need to apply on an algorithm to detect anomalies,” Chauvet said.

Using data lakes for this purpose at Deutsche Bank has proved “somewhat successful,” though this approach does not yet work on the trading floor to provide real-time analytics that detect patterns and signals in real time, or for risk, where the bank found it was easier to normalize risk across the different business divisions and then have a data warehouse. But Chauvet hasn’t given up hope: “This is an area which is evolving very quickly and the techniques are evolving very quickly. As we learn, we may be able to make those cases work,” he said.