Brown Brothers Harriman Experiments with Machine Learning

Brown Brothers Harriman (BBH), which services more than $5 trillion of assets, is developing its own machine learning models to detect price anomalies and eliminate manual processes involved in reconciliation, Michael McGovern, managing director and head of the firm’s investor services fintech offering, said while speaking on a panel at the North American Financial Information Summit (NAFIS).

Like many financial services firms, BBH is testing the machine learning waters and has found that a natural area for adoption is in the reconciliations realm. BBH’s trained and supervised model pulls in client data and combines old-school statistics with new forms of AI to “provide a signal and separate it from noise, and then deploy that as an input into a reconciliation process,” he said.

• To read about how BNY Mellon is using machine learning to make access to—and understanding of—its documents easier for its employees, click here.

As an example, machine learning helps bring efficiency to one of the products under BBH’s Infomediary franchise, called InfoRecon, which makes use of a back-end reconciliation engine. In working with one client, the data coming in was ideal, according to McGovern—clean, adequately governed, high quality. Yet, the match rate hovered around 91%. By using machine learning algorithms to better find and highlight breaks, that figure reached 98.4% by identifying additional unmatched items.

Perhaps even more surprising, the model was trained entirely over a single weekend.

“If you think about it, the difference between 91 and 100 is all the manual processing left in that particular process. With this one model, we’ve eliminated 90% of that,” he said. 

Discounting the labor involved in just collecting a massive amount of data, McGovern said the line between opportunity and challenge is a little bit blurred in regards to democratizing machine learning and integrating seamlessly across the industry. The key is to get the people equation right, first.

“Get the tools in front of the people that have the domain expertise and have analytic mindsets and expertise in a specific area—could be an asset class, could be an investment strategy, could be geography, could be some arcane form of risk,” he said. “This is not my term, but I love it, so I plagiarized it: ‘I think we need to create a generation of citizen data scientists, and enable the analysts we have in different parts of our business to utilize this toolset to generate the next set of insights.’”

• To read about how Zurich Insurance is trying out a Data-as-a-Service (DaaS) model, in conjunction with AIM Software, to be able to direct its energies toward high-level analytics and artificial intelligence, click here.

McGovern also noted that the firm has started using machine learning to detect price anomalies, but did not provide further detail.