The critical need to improve data quality and data management across capital markets will continue to escalate through 2020 as firms delve deeper into digital transformation, regulators demand more granular and ad hoc compliance data, and businesses look for much needed new business opportunities.
These use cases of data management are high on the agenda at many firms and ongoing challenges for data management teams, but help is emerging as artificial intelligence (AI) technologies mature and become increasingly important to data management as a discipline, and to financial services firms discovering that some of their biggest data management challenges cannot be solved without AI.
The call for AI technologies
Banks and other organisations within the financial services industry are working with large datasets that have to be constantly updated, monitored, and assessed. Today’s plethora of electronic data forms – both structured and unstructured – makes this task even more complex. Regulatory filings, corporate reporting, news, social media, and other data sources all need to be combed through for new information about a client. This information then needs to be combined with data from subscription services or feeds that organisations subscribe to, as well as existing internal records. Adding to the challenge is that the data – internally as well as externally – could be in a variety of languages.
Until recently, there were few options on how to manage complex data – firms had to hire people, lots of them. The use of people to monitor information channels and update databases presented its own issues. First, people make mistakes, and these mistakes create significant risks, such as compliance risk and operational risk. Second, it can be hard to find people with the right skill sets, at the right salary levels, to do this kind of work. Third, it takes people a lot of time to do this work well – to find, read and analyse each piece of content – adding to costs.
Creating new approaches to old challenges
The advent of AI techniques is changing this. For example, a large international bank might have millions of customer records in its client and counterparty master dataset, which becomes very difficult to keep up-to-date over time, says Matt Good, chief technology evangelist at Kingland, an Iowa, US based provider of enterprise data management, text analytics, and compliance management solutions.
“To maintain that data effectively with a better cost curve requires natural language processing (NLP), where data is sourced from unstructured content or documents,” says Good. Text analytics can also be applied, so that, by way of example, mergers and acquisitions can be detected and understood in the context of managing the overall dataset. Good adds: “This approach can really inform and keep up the health of a data management program related to client and counterparty lists.”
MarkLogic, provider of an operational and transactional enterprise NoSQL database platform, uses a support vector machine classifier within its machine learning (ML) technology to deliver what the company calls ‘smart mastering’. Giles Nelson, chief technology officer, financial services at MarkLogic, explains: “This combs through incoming data and identifies fields and parts of the data that it believes are similar and could have a semantically equivalent meaning. It then uses this information to combine data.” For example, the solution could identify ‘Bob’, ‘Rob’ and ‘Robert’ as the same individual and help tie together customer records from different business lines.
AI technologies are also being applied to a related problems, particularly regulatory requirements around know your customer (KYC), including initial customer due diligence and ongoing, continuous monitoring for changes in the customer’s risk level. This is an area that quickly gets into the realm of foreign languages, particularly in private banking and wealth management. Challenging languages are Russian, Arabic and Chinese, because the writing systems and structure are so different to European tongues. These languages represent a new challenge for NLP. The technology has to be able to make sense of what it is reading, identify individuals correctly and analyse the context in which the name appears to gather risk relevant information, such as whether the individual presents a financial crime risk.
Dermot Corrigan, CEO at SmartKYC, which offers AI-enabled technology to automate third-party risk intelligence collection, says: “Our expectation is that NLP will be central to how people mine data for intelligence.” Today, he says NLP, coupled with ML, can help technology learn how to identify a name in Arabic and translate it as a name, rather than transliterate it. The technology can be trained to identify the language a media is in, perhaps Chinese, Japanese, or Korean, and can also be trained to identify relationships within text, for example the difference between a judge and a convicted criminal, and two convicted criminals in a newspaper article.
Corrigan says: “This might sound like minutia, but all of these little marginal gains, all put together, mean that a team of analysts can get through a much greater workload.” Ultimately, he says, AI applied in the right way should enable people to “spend less time doing research, and spend more time making decisions.” This shift makes continuous monitoring of KYC risks a real possibility, without having to hire additional armies of analysts.
Looking into the crystal ball
So, what does the future hold for AI within financial services? Certainly, there are challenges ahead. For example, regulators want financial services firms to be able to show how the technology works, including the models used for ML and how they are being trained, says Jesse Sommerfeld, head of data science at Kingland.
Then, regulators want to see how firms are validating that new training inputs that the model is receiving over time are not degrading the model because they are less relevant. A good, if extreme, example of how this might happen is the Microsoft Tay chatbot, which was trained after launch by members of the general public to say offensive things. While tech companies can further evolve NLP and ML analytics, they must also make sure the training and activity of these technologies is auditable, including the data used. Sommerfeld says Kingland already provides this auditability for the technology it provides, but it will remain a challenge for the industry as a whole as the sophistication of models evolves.
However, the possibilities are exciting, The kind of enhanced KYC analysis SmartKYC is building could be applied to ethics, social and governance (ESG) search, says Corrigan. “I think businesses will look at what is the broader behaviour, the broader profile, of this individual or company and the ethical factors,” he says. “I would argue that all banks, in due course, will consider ethical factors in their onboarding decisions. These ethical factors will count as much as financial factors as banks are realising the impact that unethical clients can have on reputational risk.”
AI may even be able to help firms understand the questions they should be asking. Nelson says that applying a variety of different AI techniques to a large dataset could eventually help banks discover the so-called ‘unknown unknowns’ – questions they never thought of asking. For example, applied to KYC within banks, AI might eventually be able to identify patterns in client data that are red flags for criminal activity that human minds have not conceived.
In short, AI has come a long way over the past few years and there have been some exciting developments. While the future will present challenges, both anticipated and unanticipated, experts in AI believe there will be plenty of opportunity for financial firms to change the way they undertake data management.