A New Dawn for Scenario-Based Surveillance

Published On January 28, 2019
7 MINUTE READ

The era of lexicon-based trade and communications surveillance has moved past its peak as a powerful new wave of tools driven by machine learning and Big Data raise the bar for compliance and satisfy the increasingly stringent demands of regulators.

Not only have several vendors from the first generation ceased activity, financial services institutions are demanding additional solutions to address their surveillance shortcomings in light of new regulation, or to address findings by regulators.

If any one event marked the need for a new and different way to monitor and manage conduct risk, it was the $2m fine handed out to an independent broker dealer for lexicon deficiencies. The firm had a surveillance system in place, but it somehow missed crucial information for almost a decade.

The traditional lexicon-only approach, based on key words and phrases, designed to detect “previous manipulative behavior” that has come to light, is no longer adequate, and crucially will only catch inappropriate behavior that is done in the same or similar way.

The regulators have woken up, has the industry?

The shortcomings of this type of surveillance are well recognised by the users of such systems as well as the regulators, forcing compliance departments to look for ways to improve their results or find alternatives.

It is clear that firms must invest in new technology and opportunities to transform and improve their surveillance models and techniques.

Happily, there is increased willingness across the industry for market participants to jointly try and solve these types of industry issues and identify the right solutions, and Baringa is a key proponent in extending support to this approach, having formed a strategic partnership with Behavox to accelerate reform across the evolving world of scenario monitoring.

An institution revisiting their surveillance set-up will need to answer the simple question: Is scenario-based surveillance sufficient and future safe? Is there a need to look beyond the tried and tested? The answer to this question is influenced by the aim of the surveillance function, i.e. is the single goal an avoidance of regulatory findings and commercial fines, essentially box-ticking, or should this be extended to include differentiation through compliance?

There will also be significant differences in approach depending on the market activity and size of respective firms.

While large, multinational organisations might depend on separate solutions and a more finely-tuned scenario set-up, not just by location but at an individual desk level, firms that handle limited volumes or are pure price takers in the market tend to naturally opt for a less differentiated approach at the moment.

It is extremely difficult to design and improve scenarios at an individual firm level, as there are not enough data points or events (true positives) to give high levels of confidence about their effectiveness.

What are the trailblazers doing?

The surveillance industry is being reshaped by much more advanced software and alternative surveillance techniques, which combine holistic solutions across communications and trade surveillance with artificial intelligence (AI) and behavioural analytics.

These advances offer the potential to pre-emptively identify suspicious behaviour indicative of manipulation before any damage has occurred, in addition to very tangible benefits related to the perennial focus on false positive reduction driven by cost pressure.

Alert automation can be drastically increased and the industry is moving towards holistic and combined solutions, despite 80 percent of trade and communications alerts currently relying on static rules-based thresholds.

At the recent AFME London financial services compliance event, regulators were asked what concerned them the most about new technologies, and the answer was unanimous – black box solutions. Banks need to be able to convince the regulators that their scenarios address the regulations, and also allow a detailed review of how the algorithms are actually working.

Firms adopting this approach will take advantage from the rapid growth in volume and variety of scenarios, and in collaboration with peer firms contributing to the design and testing. This combined approach will create a powerful solution that will ensure appropriate and proportionate algorithms applicable to the scale, size and nature of their business and areas of interest. This granular and tailored method has never been achieved before, but offers an unparalleled level of protection for compliance and risk control.

Lexicon-based scenarios are unlikely to ever go away – users are comfortable with them and regulators accept their use. But the industry is stirring and acknowledging the fact they cannot capture absolutely everything. The solution isn’t to add more words or more people to review the output.

Lexicon-based search, used alongside techniques such as Natural Language Processing (NLP), can eliminate the majority of false positives and craft the most meaningful alerts or insights that require attention.

NLP classifiers can even be created and trained to remove content types that should not feature (such as disclaimers), or to target specific pieces of content. Pattern scenarios are also emerging as formidable ways to monitor for suspicious or unusual behaviours, which warrant further investigation. While the reason may be harmless, it could be a sign of something more concerning that has to be addressed quickly. We now know that one notorious ‘problem’ trader would leave work but then go back in later when nobody else was in the office to do his nefarious acts. A pattern scenario set to track unusual hours of monitored employees would have highlighted this behavioral anomaly relatively early on.

We’ve seen countless examples of smart individuals attempting to game the system, lexicons are ripe for this, and some even see it as a challenge to try and circumvent surveillance. But it is very hard to hide changes in behavior versus a historical/peer norm, where a model can detect subtle deviations in an employee’s behavior compared to their past habits or the standard of their peers.

A truly effective surveillance solution needs to be proactive about emerging risks and themes that encompass output from key regulators, partnerships with industry experts, research done for industry journals, and the compliance community in general. Through active engagement and exchange with the industry, the goal is to create an industry benchmark.

There is an arguable case for scenario-based surveillance and the ease of demonstrating a level of regulatory compliance to regulators and auditors. These models will consistently provide a level of assurance in avoiding most enforcement where employed correctly, and they require a reduced competence within Compliance functions to operate.

But the benefits offered by a tool that uses artificial intelligence (AI) and behavioural analytics far outweigh those of legacy approaches. These next generation solutions are likely to have a significantly higher success rate in finding the true positives, while at the same time materially reducing false positives that occupy legacy platforms.

What does the future hold?

The sun is setting on lexicon-based legacy solutions, but natural caution among the risk community means a hybrid approach is the current direction of travel. This frames a use of clear and easily explained scenarios, supplemented by statistical analytics that are combined with high-level behavioural analysis through graphical displays of relationships and data.

The future looks very different. No more blunt tools that tie compliance analysts to a screen checking endless false positives that help to appease supervisors looking to check a box and move on. Imagine a genuinely thoughtful approach to surveillance where monitoring is proportionate and targeted like a laser at the real areas of risk (be that a desk, an asset class, a data type or even a communication channel), where using a technique like NLP to extract sentiment or an entity from a data set can reveal a disaster that is in incubation but still preventable. Imagine a world of surveillance where the creativity and knowledge of the human using the tool can determine the success and effectiveness of the risk control framework of an enterprise. That is a tomorrow worth embracing.


Baringa Partners is a specialist Management Consulting firm with ~700 people in offices across the UK, Europe and the US. We are passionate about being ready for the future. The greatest strength of our diverse team is the collaboration with our clients and bringing people and organisations together to solve the biggest challenges that financial institutions, market utilities and the regulators face. We are independent and collaborative, bringing all of our internal cross-industry and capability experts, as well as our network of external partners, from policy makers to Fintechs, to every client and challenge we work on.