The rush to adopt AI is unrelenting. But can it go too far? The main push for using AI in trade surveillance is to improve efficiency and reduce the noise around the volume of false positives that still threaten to swamp compliance teams. However, some trade surveillance systems are increasingly using AI to generate explanations and the reasoning behind alerts; for example, providing firms with real-time risk analysis of news alerts and their potential impact on market movements.
While there’s no doubt that AI offers huge potential to help regulatory professionals to do their jobs more effectively, the conversation goes too far when we talk about closing out alert investigations purely through the use of AI and automation. AI-resolved cases are all non-deterministic – you don’t get the same result every time you run them. So, from a compliance and quality assurance perspective, relying on AI completely carries several major risks.
What’s more, human judgement and expertise are key ingredients in assessing cases and providing context, especially given AI’s propensity for bias and hallucinations. The technology holds great potential to assist compliance teams – but as a co-pilot, not the captain.
If it ain’t broke, don’t fix it… but do enhance it
Earlier this year, Microsoft’s CEO Satya Nadella said that up to 30% of the company’s code is now written using AI. But that’s only for certain programming languages, mainly Python. As coding effectiveness varies for other languages, should there be concerns around its growing use? What it does highlight is the importance of human oversight in assessing where and how AI is utilised.
If this was replicated in the compliance world, trade surveillance systems might lack the performance needed for AI to tangibly enhance processes. Any poorly coded systems could lead to AI models colluding with each other and machine reasoning differing from human intent – risks very much present in trading systems using AI. If a firm had programmed an AI model to process alerts with a focus on speed over accuracy, then this could lead to instances of suspicious activity not being flagged or missed entirely.
AI works best when it is integrated into an already highly-effective system. The most advanced trade surveillance platforms have a wealth of data and use their own work rules and logic without the use of any external services like AWS. They integrate trade, eComms and behavioural data into one location to enable a holistic approach to trade surveillance.
These proprietary models allow firms to traverse that data, assess it statistically, and easily identify any anomalies and inconsistencies (for example, when you’re generating large volumes of alerts you shouldn’t be). So, there’s a lot that can be done straight off the bat – AI can then enhance a well-established database and risk scoring process to triage alerts more accurately and efficiently.
Where AI can act as a copilot
In a regulatory context, I believe that AI is at its best acting as a copilot – a companion that interacts with your data and provides additional context for compliance teams. This approach could be used to dynamically identify any nuances in trading behaviour or communications, nuances that are often missed by traditional systems that use static thresholds for detecting suspicious behaviour. For example, defining behavioural pattern recognition for individual traders or clients, instead of applying a one-size-fits-all threshold.
By assessing contextual factors like trader history, news and market activity, AI can suggest dynamic risk scores to compliance teams – and these scores adapt as context or behaviour does to provide ongoing prioritisation. Crucially, AI’s usefulness is in escalating high-value alerts to relevant team members and not auto-closing alerts – this copilot role provides value, whereas the latter creates risk.
As such, the dynamic capabilities of AI could significantly reduce the number of false positive alerts and empower compliance teams to handle higher alert volumes more productively and with less stress. Ultimately, this gives teams the ability to more accurately spot and prevent market abuse.
A supplement to compliance professionals
While AI’s use will increase in explaining alerts and adding context to STORs, its current capabilities mean that it shouldn’t be relied on exclusively to process alerts without any human intervention. In practical terms, if an instance of market abuse was not flagged by AI or it closed an alert without investigation, it’s the regulated firm that retains regulatory responsibility for that decision. The firm can’t use the fact that AI decided that certain activity didn’t need to be investigated as an excuse – regulators simply won’t accept that.
Data quality and rules-based processes within trade surveillance systems are already vital to the insights that they generate for compliance professionals. AI presents an opportunity for firms to enhance this hybrid approach still further by supplementing decision making, not replacing it. If humans monitor and assess how they use AI as a copilot, they can build responsible automation that balances efficiency with human judgement. This will enable them to extract the most value from AI.
When it comes to the systems themselves, context engineers are key to the whole exercise – establishing alignment of priorities and goals on AI use and output are integral to its effectiveness. That’s why using advanced trade surveillance systems that integrate various trade data and provide dynamic thresholds is crucial to unlocking its potential.
One great way to test and refine new AI-generated alert configurations is through sandbox environments. These controlled and no-risk environments allow teams to fully test and validate configurations before they are deployed in the live system.
A compliance team player
There are a lot of promises being made about AI, and a lot of ‘AI washing’ – we need to take stock and be realistic about what the technology is capable of, how it can add value, and what is too risky to hand over completely at this stage.
AI can be a major asset as a supplementary tool for regulatory teams: reducing false positives, dynamically risk scoring cases, and analysing existing trade data. But trade surveillance is a particularly complex area that still requires human expertise. The underlying fact is that AI should be used to support, and not replace, the human element of trade surveillance. Used correctly, it can absolutely become an important member of a firm’s compliance team – but it’s not ready to be the captain.
