Firms are investing in new solutions for monitoring the front office in lockdown conditions, but the latest technologies raise concerns about privacy and intrusion.
The following is an excerpt from Josephine Gallagher’s article, which was originally published on WatersTechnology. You can see the full article here.
A New Frontier: Behavioral Monitoring
Traditionally, financial services firms have detected market being discussed in voice or chat communications using a lexicon, a list of words relating to bad behavior. But in cases like Libor fixing, traders just avoided using those words to trick the system, or took their conversations offline. The solutions that emerged after this scandal and others include surveillance technologies that used natural language processing and machine learning to identify unusual behavior and understand trader sentiment.
Digital Reasoning captures data from emails, chat applications, and voice calls. It transcribes voice to text, and the machine learning technology is trained to detect language that could be connected to market abuse and understand the context that frames it. The vendor also looks for unusual behavior patterns that could indicate secrecy or collusion.
Tim Estes, executive chairman and co-CEO of Digital Reasoning, says there are three important indicators to look out for when monitoring trader behavior: signs of secrecy, business language, and the parties changing their communications venue.
“If someone is trying to not be detected because they are talking about something sensitive, that is not wrong. Sensitive could be a personal thing, sensitive could be a professional thing. Once it is a business thing, with business deal language … and they get a little bit concerned and decide they want to take it off-channel, like through WhatsApp, because it’s too sensitive [for the monitored channel], when you have those three behaviors together, that is a real warning sign,” Estes says.
Digital Reasoning built a tool called Cognition, which is used to fast-track a firm’s ability to train machine learning models. The tool uses active learning and interfaces to allow analysts or compliance teams to input training data, such as examples of unusual behavior, without the need for data science skills.
One example is pressure language, when sales teams become too aggressive with clients. Digital Reasoning’s training tool has been used to detect and pinpoint signs of this happening.