CompatibL AI: at the forefront of change within the financial industry

The ongoing revolution in artificial intelligence (AI) offers tangible benefits in risk management and financial trading. By leveraging CompatibL AI, institutions can overcome the limits of large language models (LLMs) and gain accuracy, efficiency and better handling of natural language documents, while maintaining human control

The AI revolution

Over the past year and a half since ChatGPT first took the world by storm, we have been amused, impressed and assisted by the clever responses of this revolutionary technology. However, one thing that rapidly became clear was that LLMs – the foundation of modern AI – are considerably less impressive when faced with tasks that require specialised jargon or deep insight into complex concepts.

This is hardly a surprise. These models learn primarily from the information available on the internet, where texts about complex trading and risk management concepts are relatively uncommon. As a result, the knowledge of financial markets is not well represented in AI training datasets. This makes out-of-the-box LLMs less effective for financial markets application.

More from this special report

LLM integration: Special report 2024 

CompatibL knows that AI-based enterprise trading and risk management solutions would require considerable investment in research and development (R&D). Yet the potential benefits were so great that, within two months of the release of ChatGPT, it allocated a significant share of its R&D budget to harness AI for enterprise capital markets application. This work resulted in CompatibL AI – the organisation’s platform for building LLM-driven solutions for trading, risk management and quantitative finance.

Since its introduction in early 2023, CompatibL AI has won two prestigious industry awards – the Risk Markets Technology Award for Best modelling innovation and the WatersTechnology American Financial Technology Award for Best new technology (AI and machine learning).

CompatibL’s research

Alongside electronic trading and sophisticated analytics, capital markets businesses use a surprising amount of unstructured natural language documents. These documents include trade confirmations, term sheets, offering memorandums, regulatory guidelines and many other natural language document types. Quantitative modelling also produces a large number of natural language documents as part of the model governance process.

In the absence of an industry-standard format, the style and structure of these documents vary widely from one firm to the next, and even between departments of the same firm. Business processes involving these documents have largely resisted automation, and must be performed manually by teams of highly qualified experts.

CompatibL recognised that the emergence of modern LLMs created an opportunity to automate comprehension, validation and generation of these complex natural language documents.

CompatibL’s R&D team developed novel prompt-engineering and fine-tuning techniques to achieve optimal LLM performance for each document category. These innovations enabled CompatibL AI to assist with time‑consuming and tedious tasks in trading, risk management and quant research to a higher level of accuracy and competence compared to out-of-the-box LLMs.

Data security

Data security is a critical requirement for deploying AI in the financial industry – especially when processing sensitive internal documents. For an AI solution to be successful, it must offer enterprise-grade security to clients. This often requires working with clients’ existing and fully vetted cloud partner or deploying software on-premises. CompatibL offers both options for its AI solutions.

The partnership between Microsoft and OpenAI makes it possible to access generative pre-trained transformer – known as GPT – models inside banks’ Azure clouds.

Banks’ data and content are subject to the same data protection policies as any other data in Azure, and are never sent to OpenAI or other third parties. For an even greater level of security, the Llama 2 model family provides an open-source alternative – completely within the banks’ control – that can be deployed by its own engineers on-premises or in a private cloud infrastructure.

Humans in control

Unsupervised AI can’t – and shouldn’t – replace humans in performing tasks related to trading or risk management. Instead, CompatibL AI works as a co‑pilot alongside human analysts to make their jobs less tedious and more rewarding.

With CompatibL AI, humans always remain in control and sign off on the final product. An independent case study commissioned by a CompatibL client indicated that, with CompatibL AI performing the bulk of the work as a co-pilot supervised by human analysts, the overall time to perform common tasks such as trade entry was reduced by two to four times, while the accuracy was also improved.

AI will become increasingly competent over time and will be able to perform more complex tasks with less supervision. However, despite these expected advances, CompatibL’s approach will always rely on humans in the driving seat.

What-if analysis use case

What-if analysis is a critical and time-sensitive business function where the risk and capital requirements of a potential trade are analysed before it is executed. CompatibL AI uses the unique capabilities of LLMs to assist in and partially automate this important process.

While what-if analysis may not seem a natural use case for LLMs, the most time‑consuming and unreliable part of the process is trade entry from natural language trade specifications. Sometimes these specifications are obtained from term sheets provided by counterparties, and sometimes from emails or long email threads. Another important trade entry pathway is voice: performing trade entry based on verbal instructions.

While out-of-the-box LLMs can easily perform everyday tasks, none can provide the specialised comprehension suitable for trade entry. Only after models were enhanced through specialised prompts and then fine-tuned on a carefully curated set of documents with human feedback were they able to perform this task reliably.

CompatibL AI can create a what-if trade from a complex term sheet or a long email thread, correctly choosing from hundreds of options. It can also locate an existing trade in the database based on natural language instructions and apply changes specified by the user to create a new trade. One of CompatibL’s clients reported that, for what-if analysis, CompatibL AI performed well in production and significantly cut the time required to enter a new trade.

Model governance use case

Model governance is a critical bank function that can only be performed by highly qualified teams. Its objectives are to exercise control over the model development process and lifecycle, ensure all model changes are authorised and properly implemented, document the models for management and regulators, keep complete and up‑to-date regulatory and internal model documentation, and publish accurate and comprehensive model release notes.

Each regulatory submission or internal models approval requires tens of documents that often reach 300–500 pages in length. Other than the effort required, the sheer amount of information in these documents increases the risk of inadvertent omissions and errors.

CompatibL AI uses LLMs to process and integrate information from model specifications, model test results, model revision history messages, regulatory guidelines and existing documents. It can look at every line of source code and every version control log message, perform an in-depth analysis of the prior documentation and release notes, as well as integrate all of this data in a nuanced and sophisticated way.

Feedback from CompatibL’s clients has been very encouraging. During production use, its software was able to identify gaps and required changes in model documents and release notes, and generate document amendments for approval by human analysts. CompatibL AI significantly improved the quality of the model governance process, while making the work of model governance analysts less repetitive and more fulfilling.

Conclusion

From the days of using hand signals to execute trades on an exchange floor, trading and risk management have come a long way towards electronic execution and process automation. Yet, for many financial instruments, it remains stubbornly manual and reliant on natural language documents.

CompatibL’s clients see its offering as a hugely important step towards greater reliability and efficiency in dealing with natural language documents within trading and risk. CompatibL is proud to be at the forefront of this revolutionary change in the financial industry brought about by the emergence of AI.

 

Benefits of using CompatibL AI for what-if analysis

• Perform trade capture for what-if analysis from:
  • Email chains
  • Term sheets in text or PDF format
  • Unstructured data
  • Voice instructions
• Configure risk calculations based on natural language rules and regulatory guidelines.
• Use co-pilot and chat modes to provide additional instructions or apply changes.

 

Benefits of using CompatibL AI for model governance 

• Exercise control over the model lifecycle and source code at the level of detail that would be impractical without AI assistance.
• Verify code changes are correctly described in Jira tickets and Git revision messages.
• Keep regulatory and internal models documentation in sync with code changes
• Generate meaningful and comprehensive release note summaries.

Sponsored content