SteelEye, the integrated surveillance solutions provider, has successfully incorporated ChatGPT 4 into its comprehensive compliance platform. The move comes as part of a case study aimed at assessing the AI tool’s potential in market surveillance applications. The case study findings indicate that, when implemented correctly and used judiciously, ChatGPT can substantially benefit surveillance investigations.
By supporting decision-making processes, the promise of ChatGPT is to empower compliance teams to analyse data more rapidly and identify potential risks more effectively. SteelEye has deployed ChatGPT’s Large Language Model (LLM) in its UAT environment, to analyse specific communication records, such as voice calls, chats, meetings, and emails, and address several key questions. It then provides insights in the form of a content summary, an analysis of entities involved in the conversation, and a recommendation of next steps, and serves as a starting point for launching surveillance investigations and standardising workflow processes, ultimately increasing case throughput and consistency.The integration of OpenAI’s ChatGPT into SteelEye’s integrated surveillance platform signals a new era for the financial industry, although as firms increasingly request machine learning and AI tools to streamline their operations, many are unsure of the tangible benefits, suggests Brian Lynch, President of SteelEye Americas, in an exclusive interview with TradingTech Insight.
“We’ve found that firms are asking for machine learning and AI but don’t necessarily know why or have a clear vision of how it’s going to help them; there’s a general buzz and excitement around machine learning without a strong sense of what it will deliver or how difficult it will be to get true value out of it,” says Lynch.
He continues: “We built SteelEye on traditional foundations like rule-based and pattern-based algorithms for trade surveillance, and we’ve been building our own AI models and using them in small ways in our platform. Of the three aspects to providing accurate, efficient surveillance routines – content, rules, and context – we’ve found that the area where it has been most fruitful to apply practical machine learning is content. However, that’s not very visible or attractive. People want to see machine learning in action, and that’s where ChatGPT comes in. We integrated ChatGPT as a rapid proof of concept to put something visible and interesting in people’s hands and give us a chance to work with it.”
Although the technology is still in its early stages, SteelEye’s customers can now utilise ChatGPT 4 in a live environment. The model’s outputs are designed to provide contextual information, rather than definitive answers, allowing compliance officers to use their judgement and expertise alongside AI-generated insights.
“In the long run, AI might replace people in various roles, but in the short term, the context and value it offers are incredible,” says Lynch. “The depth of knowledge and speed at which AI can process data make it an essential supporting tool for anyone processing data or needing contextual elements in their work. We proved this by integrating ChatGPT and asking it nine different compliance-related questions about electronic communications. The responses were phenomenal, and there’s no doubt that AI will become an important tool for compliance officers moving forward.”
One significant advantage of ChatGPT is its ability to handle communications in multiple languages. In SteelEye’s implementation, the model analyses documents in their native language and provides insights in English. “We’ve spoken to compliance officers in multinational banks who use Google Translate to understand foreign language communications, which can be dangerous due to the loss of context,” suggests Lynch. “ChatGPT analyses documents in their native language and provides additional context that can be helpful for firms operating across multiple jurisdictions. This approach ensures consistency in applying logic across the board.”
As SteelEye expands its AI integration, compliance officers can expect their workflows to change. The focus will shift towards improving the efficiency and effectiveness of compliance tools, with AI models like ChatGPT playing a crucial role in guiding officers’ attention to high-priority issues. As the integration deepens, the questions posed to ChatGPT are expected to grow more complex, allowing for a more comprehensive analysis of multiple communications, phone calls, and trading activities.
“We’re not using ChatGPT as the primary identifier of incidents or alerts; we’re using it more for efficiency to help compliance officers decide where to apply their attention,” says Lynch. “In a real compliance case, you’re not analysing just one email; you’re analysing a series of emails, phone calls with transcriptions, and trading activity. We’ll only improve the efficiency of that process as we build on the integration.”
In the coming years, SteelEye plans to further harness AI and machine learning technologies, focusing on behavioural analytics and context. The company aims to provide additional tools to help compliance officers examine behavioural patterns and changes over time. As a result, AI-driven technologies like ChatGPT will become an indispensable part of detecting anomalies and improving compliance processes, suggests Lynch.
“We’ve got content curation well done and will continue using machine learning to avoid false positives by weeding out irrelevant content,” says Lynch. “We’ll also focus on context, as compliance officers want to look at the behaviour around primary events, changes in behaviour over time, and whether that indicates more attention should be paid. ChatGPT has shown it can be helpful in detecting anomalies. It’s an almost endless journey, and as we get more comfortable with AI, we’ll continue focusing on the data coming from compliance teams to produce helpful results from an efficiency perspective going forward.”
The widespread adoption of AI tools in the financial industry does raise questions about the evolving regulatory landscape, notes Lynch. One ongoing debate concerns the balance between effectiveness and explainability. Although AI models offer powerful capabilities, regulators still require traceability and proof of their effectiveness.
“Proponents of older, rules-based models focus on explainability,” he says. “When a regulator comes in, you need to provide traceability, not just say you have a robust AI framework. We must prove the effectiveness of these models, and once we do, the traceability matrix for regulators will evolve. This will allow machine learning to be more widely accepted in the regulatory landscape.”
Subscribe to our newsletter
