AI Chat Surveillance for an Investment Bank

ai data concept

Our data scientists have created an AI surveillance solution for Bloomberg’s IM service, Instant Bloomberg [IB], for a global investment bank.

Banks are reluctant to let large organizations develop models based on their data.

Context

Today’s trading floor communications take place across multiple systems and modalities. Voice remains a primary channel, but the bulk of communication takes place on instant messaging services. MiFID II Article 16 (6) requires records to be kept of all services, activities and transactions including all communications that are intended to result in a trade even if they ultimately do not.  Article 16 (7) requires recording and record keeping of telephone conversations or electronic communications. In all areas of communications regulation, surveillance follows recording and storage. What’s more, surveillance demonstrates that every effort is being taken to control risk. So, while the primary focus of surveillance regulation is on the trade data, proactive supervision of all communications channels used leading up to a trade is important for trading organizations.

Client Requirement

Our client wanted to enhance its surveillance capability by monitoring communications on this key order management channel. The primary objective was to identify traders’ deal intention when discussing trades on IB. This would enable the fulfilment of regulatory requirements to report all communications containing deal intents using a ‘Request for Quote’ (RFQ) format while facilitating monitoring with a much shorter lead time.

Our Solution

Banks are keen to harness the analytical power of AI but are justifiably reluctant to let large organizations develop models based on their data. This makes it logical for banks to engage specialist consultancies like ours when developing data-focused solutions. Our scientists have access to the most advanced Natural Language Processing models (NLP) used to analyze text data, and the bank’s solution remains unique and protected. With completed projects like the one described here, we have the expertise and experience to replicate the solution quickly, accelerating the process despite starting with unpopulated models.

Our team collaborated with the client to prepare the IB NLP solution. We established the surveillance requirement and the specific reporting outputs needed to optimize governance. The initial familiarization period involved learning the language and nuances of the trade lifecycle across different asset classes including Equities, Forex, Fixed Income and Derivatives. As with our TIMM™ service, a number of databases were interrogated to enhance the analysis, improve confidence levels and avoid false findings. The key data source was the Bloomberg system, which generated 10GB of downloads per week for this client.

Different names, identifiers, slang and jargon were established for products, desks, and trade activities. We did the same for individual people inside and outside the bank. Even in a corporate environment, chatroom IDs can be quite informal. The solution had to be precise about who was the sender and receiver, employee or client in any given conversation. This deceptively simple work provides the foundation for accurate surveillance, allowing our data scientists to build appropriate data frames.

The NLP solution automatically extracts and compiles a daily report, summarizing all deal intents expressed on IB. This ensures comprehensive analysis and accurate insights are achieved. In order to support thorough oversight and investigation, the report presents the true intent and complete product information, determined through the utilization of advanced machine learning algorithms [ML]. The main report fields were:

  • True Deal Intents
  • Financial Products Mentioned
  • Distinguishing the Unreported and Mass Data
  • Flagging the Responsible Person

Outcome

Our solutions met and exceeded the client’s expectations. The client was able to reduce its exposure to regulatory risk while enhancing its surveillance operations. The solution provided data insights to be drawn upon in the development of the bank’s subsequent surveillance and regulatory tools. It also advances our own capabilities allowing us to provide more sophisticated services to new and existing clients.