The capabilities of Large Language Models (LLMs), like OpenAI’s ChatGPT, are undeniable – and maturing rapidly. With the recent launch of custom GPTs, developers can now build LLMs for the specific needs of their business. Even so, valid concerns remain in the heavily regulated financial services industry.
Financial services firms can judiciously embrace the possibilities. The key to leveraging this technology in a timely but prudent way lies in customising and privately deploying LLMs to ensure compliance with strict privacy, data handling, and regulatory requirements.
For banking and finance institutions in Hong Kong and Singapore, three critical pain points have been identified which LLMs can effectively address.
Diverse and complex data analysis and reporting
Data analysis and reporting can be challenging for financial institutions managing dozens of lines of business across multiple dynamic markets. Modifying the parameters of Business Intelligence (BI) tools for different tailored reports and diverse business operations is complicated and technically challenging; business users must often rely on the technical team even for simple queries.
A conversational BI solution can revolutionise this process. Integrating a privately deployed LLM with existing BI tools allows business users to query and visualise data using natural language, making data analysis more accessible and efficient. This process demands extensive training of the LLM with a wide range of datasets relevant to varied business operations.
The LLM will need a comprehensive understanding of the various business domains, as well as the ability to interpret speech and text inputs, convert these to complex data queries, and then generate useful insights.
Time-consuming contract reviews
In commercial banking, time-consuming manual reviews of contracts are necessary to facilitate trade finance, bill discounting, mortgages and loans, and other services. LLMs can expedite this process by automatically extracting key information and comparing it with internal business rules and regulations.
Training LLMs for this task necessitates inputting a diverse array of contracts, internal guidelines, and workflow processes. The model must learn to recognise and interpret complex legal terminology, contractual obligations, and conditions specific to different contract types.
Over time, it will be able to identify key contract elements and compare them against the bank's compliance standards and operational procedures, highlighting any discrepancies for further review by a human and saving significant time and resources.
Recent moves to combine the power of LLMs with visual data processing capabilities are making it possible to automate the extraction of contract and invoice details from a broader range of scanned documents and images, further streamlining the review process.
Realistic test case and test data generation
In software and system testing, creating relevant and effective test cases and test data is vital but time-consuming. This is particularly true in the finance and banking sector, where accuracy and compliance with regulatory standards are paramount.
Test cases are usually designed by experienced testing professionals based on business requirements, while test data must be carefully prepared and maintained to align with intricate business rules and realistic scenarios.
Applying LLMs in testing requires locally deploying an LLM that has been pre-trained in banking and financial knowledge. The LLM should then be fine-tuned with business processes, existing test cases, and test data to allow it to generate appropriate test cases based on natural language inputs directly from the business team.
The LLM may even help generate test data that reflects real-world business scenarios – simulating customer transactions, account behaviours, and market fluctuations to ensure that the testing environment closely mirrors actual operations.
LLMs already offer ground-breaking opportunities for innovation and growth. But integrating LLMs into the highly regulated and sensitive banking and finance sector demands a strategic approach. To ensure successful adoption and seamless integration, we propose four key steps:
- Collaboration with experts capable of customising and implementing generative AI solutions to address the institution’s unique needs and challenges.
- Private, customised deployment to ensure alignment with the institution's specific data protection policies and regulatory requirements.
- Continuous fine-tuning post-deployment to tailor the LLM to specific operational needs and business use cases.
- Regular updates and retraining sessions to ensure continuous learning, compliance with the latest banking and finance regulations, and adaptability to new financial contexts.
From enhancing efficiency in BI to streamlining contract review and revolutionising quality assurance, the potential applications of LLMs in the banking and finance industry are vast and transformative. As the technology continues to evolve, so too will its impact on the industry.
However, this journey requires careful planning, with a focus on data security, regulatory compliance, and customised solutions.
