NICE Actimize, the US software giant, is now marketing a trade-related surveillance programme that, it claims, 'ensures' that user-firms will comply with MIFID II, the Dodd-Frank Act and Regulation Best Interest.

The software, known as SURVEIL-X, analyses trading data that it uploads from user-institutions to the Cloud. Indeed, NICE Actimize says that it is the first software on the compliance market to do this. It claims to detect previously undetectable scenarios or types of behaviour that software would not normally spot if it ran on 'rules only' conditions that looked for known scenarios.

The present generation of trader-surveillance software looks at batches of trades, or a sample of them, and then follows rules laid down by its programmers to evaluate them. The new software looks at all the available data and uses artificial intelligence to analyse it all at once.

When AI analyses this data it detects anomalies by looking at anything and everything to do with trade-related behaviour and establishes mathematical relationships between the various events. It then spots moments when things happen that do not fit in with the relationships that it has established. As one spokeswoman for NICE Actimize put it, "it looks for dodgy behaviour by looking for patterns and when they don't match up, it generates an alert." As another spokesman put it, "in order to detect these types of unknown scenarios, we are using unsupervised machine-learning techniques, specifically anomaly detection, in order to highlight behaviours that do not fit the normal patterns."

The software also claims to analyse speech and behaviour better than previous examples. Another technique that it uses is Natural Language Understanding, which looks at the text of emails and discerns its meaning. When software detects anomalies, it often creates many false positives; Natural Language Understanding then comes in to trim away most of the 'noise.'

Another technique is Smart Classification. To explain this term, the spokesman told Compliance Matters: "We are using Supervised Machine Learning techniques and Natural Language Processing in order to classify content based on what the model has learned from the participation of users. This is ultimately used to reduce the false positives, but also to enable investigators and analysts to identify needed content in a smarter way."