FCA Using NLP, Machine Learning to Regulate Businesses

The Financial Conduct Authority (FCA) is using emerging technology like machine learning and natural language processing, as part of its broader strategy to utilize data more effectively, said Steven Green, who is head of central data services, part of the innovation division in strategy and competition at the regulator.

“We are looking at natural language processing for analyzing documents, business plans—we have a lot of those materials in the organization. We are using machine learning to help us predict harm, to look at those areas of the market that we should be focused on,” said Green.

Green was delivering a keynote address during the TSAM Digital London conference. 

“We are looking at combinations of new datasets to spot outlier firms, to look at patterns of firm’s behavior, to look at the way the data reflects those firms that act differently to others, to see what’s going on there, and maybe that allows us to focus our efforts when we have such a broad suite of firms we are looking after,” Green said.

The FCA is also looking at graph analytics, which uses algorithms to understand the relationship between entities on a graph, to track criminal networks and gather intelligence on social media and the internet at large.

“A lot of what we do is about understanding where the outliers are, where the individual firms or people are that are doing things that we want to understand better, or that we don’t like,” Green said.

He said the FCA has been thinking about what data it needs within the organization, as part of a years-long initiative to position itself as a digital regulator. This is more and more important, especially as the universe of firms that it supervises is now some 62,000 companies.

“[Understanding the outliers] is a bit easier when you have a small number of firms, the ability to process data quickly, and being able to target your interventions. That naturally becomes harder to spot as more and more firms exist in your remit, it becomes harder where the patterns of analysis we are trying to get after are more complex, and if you combine that situation with large amounts of data required to understand those patterns, to be able to spot those outliers, it becomes very difficult to do that with techniques that existed five years ago,” Green said.

Holistic Data Strategy

About three years ago, the FCA established an advanced analytics unit to look at individual use cases at the front line of the organization where it could use data more effectively. It then set up pilots to test these cases. At the same time, it ran a workshop consulting teams on the ground on their ideas and frustrations, and to learn about ways of utilizing data differently.  

“A lot of these good ideas were about process change, so this is not just about delivering data and tech, it is about a holistic view of how we change an organization by utilizing those capabilities and changing the way we think, the way we work, the way our processes are set up,” Green said.

The FCA is now looking to expand those use cases across the whole organization with a holistic data strategy. Although it has policies and controls around data quality and so on, as part of its traditional data strategy, the FCA wants to gain a deeper understanding of how markets work, how consumers behave, and to be flexible for an uncertain future. “We are also reflecting on our culture and skills more broadly: the ways of working, the leadership and judgment-making that we need in place to utilize data and technology more effectively,” Green said.

The strategy focuses on utilizing data for maximizing insight, becoming more effective in using data by removing manual processes, and by activating expertise in the organization.

“This is not just about creating data science teams, it’s about making sure the whole organization is clear and comfortable with its use of data, and working with our IT colleagues to power analytics and build new technology stacks,” Green said.

The approach is underpinned by data science teams scattered through the organization, with one central team that coordinates activities. This hub-and-spoke model intends to encourage staff to take ownership of projects and datasets, rather than feeling that they are working to fulfill edicts from on high.

Scalable

The FCA has been migrating to the cloud for the past five or six years. Green said that by the middle of next year, most of the regulator’s services will be on a new, scalable estate “that allows us to gain the benefits of new technology.”

The FCA is also looking at expanding its data sources. “We don’t just gather regulatory data. We gather data from charities, from other government agencies, we purchase data where possible off the market. We are looking at how we can be more flexible with the sourcing we do when it comes to the acquisition of data,” Green said.

Gabriel, its online system for collecting and storing regulatory data reported by firms, is being replaced over the next year. The regulator is also in the process of building a new data lake. “We have notification systems that are being built and replaced that help us do automatic case creation, that look at monitoring our datasets over the lake, and help us understand our interventions,” he said.

On top of that, the FCA is looking into how it can safely share data with the market, and how it can industrialize the analytics tools it has developed for its data science teams so that the market can use them.

It is also trying to attract new talent from the market with programs including a data science career program, launched last year. According to Green, this initiative attracted 2,000 applications for 10 roles.