AI Surveillance Trends:
Three Experts Weigh in on Why AI’s Essential and What’s Coming Next

According to recent survey by Ernst & Young, AI adoption among financial services firms has become widespread with 99 percent of firms saying that they are already deploying AI, and all citing that they are either already using or planning to use generative AI (GenAI). The majority of respondents felt optimistic about the benefits GenAI would bring to their organizations, with long-term sentiment registering even stronger. While the Ernst & Young survey covered broad uses of AI across financial services organizations, a separate Chartis Research Survey (The Future of Trader Surveillance) revealed that financial services firms are also prioritizing AI to improve trade surveillance, with half of firms surveyed listing AI as a key technology driver. Another NICE Actimize survey revealed that 25 percent of respondents were actively using Natural Language Processing (a form of AI) for communications surveillance. With these trends in mind, we asked three AI and regulatory subject matter experts to weigh in on how they expect “AI will improve surveillance for financial services firms in 2024 and beyond.” Here’s what they had to say:

AI is transforming how companies are using surveillance to control regulatory risk. Specifically, it is helping  firms improve the efficiency of human resources engaged in surveillance, by:

• Prioritizing the work queue of analysts so they can allocate time to what matters most. For example, AI can learn from previous user feedback to score and propose relevant actions to take. This allows compliance staff to focus on the most important alerts first.
  
• Providing explanations for alert results. For example, AI can learn from existing rules to highlight relevant portions of input data and summarize results in plain English, allowing users to allocate more time to relevant alerts.
  
• Analyzing increasing volumes of transactions and identifying anomalies. AI can also leverage trade and alert data to detect deviations and help shine the spotlight on entities that might have otherwise been missed due to being just below thresholds.

The growing use of AI doesn’t mean that firms are abandoning rules-based analytics. Regulators are very prescriptive in terms of what constitutes bad behavior, so there will always be an important role for rules- based analytics (that don’t require AI). We are also starting to see firms deploy rules-based analytics in conjunction with AI, where the AI can learn from existing rules as well as user feedback to improve the analytics. I believe in the long-run that firms will benefit from this co-existence of AI and rules-based analytics.

One of the biggest areas where AI is having an impact, and will have the largest future impact, is in the area of communications surveillance. Voice communications is unstructured data. Understanding communications data begins with AI. First you need to accurately convert the audio to text in whatever language it’s in. Then you need to understand the context of the communication. What is the sentiment being expressed? What is being said? If it’s an email, you need to know if it’s a business email, or spam? Does it have a disclaimer? Financial services firms can reduce false positive alerts, improve efficiency, and save time and resources by using AI and NLP to:

• Improve transcription accuracy;
• Accurately detect sentiment in conversations;
• Weed out disclaimers and non-business communications;
• Tie trades and unstructured communications together and reconstruct events (using AI along with entity extraction).

When it comes to communications surveillance, AI is a must-have.

Paul Cottee, Director, Financial Markets Compliance

As trading volumes and business complexity have increased on many markets in recent years, the surveillance workload generated by this increased activity has also grown. However, the capacity of surveillance teams to manage this workload has not kept pace. Surveillance managers are all looking for ways to improve efficiency. AI can assist in many ways. Here are a few examples:

• False-positive reduction: AI processes can carry out basic investigative work, and determine if alerted scenarios are more likely to be activities which are benign or understood by the business, and therefore unlikely to be suspicious.
  
• Alert scoring: when faced with high numbers of alerts, it can be difficult for analysts to prioritize their workloads. AI can learn from previous, similar activities, and use this to assign confidence scores to subsequent alerts, which can in turn help analysts prioritize which alerts to focus on first.
• Auto-closing: AI can learn the disposition of common scenarios, and based on what it learns, identify and automatically close benign alerts. Likewise, it can learn from the disposition of true alerts, which cases are likely to be true positives, and even go as far as drafting the appropriate paperwork for escalation or regulatory reporting.
  
• Management information: AI can be used to automatically generate reports for management. This might be as straightforward as generation of regularly-scheduled reports, or reports detailing where a particular variable has changed by a noticeable degree since the last reporting point. Again, the AI might take over the task of drafting the report, meaning that the analyst’s time can be better spent on investigating true positives.

It should be noted in all the above scenarios that human supervision will still be required. While regulators are encouraging the adoption of new technology, they still emphasize and expect ‘explainability.’ There will aways be a requirement to be able to understand and explain an outcome. This implies that humans will still need to review any AI output.

Nitin Vats, Product Manager, NICE Actimize

In years past, financial services firms performed communications surveillance and trade surveillance in silos. As regulations continue to evolve, firms are now embracing holistic surveillance. You can’t have holistic surveillance without AI. True holistic surveillance requires the ability to efficiently analyze large volumes of trading, market, and communication data. Analyzing structured trading and market data is fairly straightforward with rules-based techniques. However, analyzing unstructured data related to employees’ communications (particularly voice) requires modern AI-based techniques to truly understand the content and the context of communications.

In addition to this current use of AI (described above), here are some future uses of AI in holistic surveillance that I believe are also worth exploring:

• AI-powered Co-Relation: Better surveillance relies upon the co-relation of communication and trading activity. Generative AI can enable compliance analysts to find the related information for a particular trade with just a few prompts, simply by embedding a few prompts. For example, prompt one could be, “Create a transcript of all Microsoft stock-related interactions on February 8, 2024, between Steve and Alan.” Additional prompts like, “Highlight sentences from the earlier retrieved conversations that have the intention of hiding information” could be used to further refine the results. This would eliminate a lot of manual work and time compliance teams spend chasing down relevant information.
  
• Recommendation Learning to Expedite the Investigation Process: Compliance teams often spend an exorbitant amount of time navigating through tons of alerts. This is especially true for junior analysts who are not as skilled as senior analysts. Recommendation machine learning can help junior analysts improve their efficiency by providing junior analysts with a list of steps and best actions to take, based on specific alerts they are reviewing. For example, the system might recommend the following actions to the junior analyst:
  

  a) Consider reviewing the communication data for this trading alert.

  b) For this type of alert, other analysts usually spend more time reviewing unusual trading patterns, such as large purchases or sales.

• AI-Based Reporting: Today’s rules-based reporting is limited to what is defined in the rules and doesn’t provide insight into anomalies or trending information. With AI-based reporting, managers would be able to see surveillance trends and outliers that might otherwise be overlooked by rules-based reporting. For example, an AI-based report might reveal that a certain number of traders on a particular trading desk discussed a specific stock, and the trading volume of that stock subsequently increased by a certain percentage. AI-based reporting would give managers insight into hidden anomalies or outliers in data.

While AI holds a lot of promise in the surveillance domain, financial services firms would be wise to focus on the following to ensure AI implementation success.

• Model Governance and Monitoring: Managers should pay attention to how the models are performing, and whether the models purport a potential bias for any particular group of people or location.
• Cost in implementing Generative AI: Traditional machine learning and NLP-based models can be hosted without spending a lot of money. However, generative AI is relatively new, and hosting these models, in particular making use of APIs like Chat GPT, can be quite expensive, and this can eat into profit margins for firms.
• Use AI and Humans Together: The adoption of AI can only be successful when humans invest the time and effort to make AI better and more reliable. Your compliance analysts need to spend considerable time reviewing the decisions made by the AI, and continuously teaching the models. Finally, your compliance analysts should never make blind decisions based on AI. Rather, they should rely on it incrementally, as they build trust in the models over time.

Learn more about AI’s impact on compliance and surveillance:

White Paper: Smarter Communications Surveillance with AI

eBook: AI-Assisted Alert Reduction

In-Focus Report – Solving Surveillance: Detecting Market Abuse & Conduct Risk Across Modern Communication Channels