Applications of Generative AI in Finance

Summary: What is Generative AI?

Generative AI – a machine learning model capable of generating new data in response to inputted prompts through Generative Adversarial Networks (GANs). It primarily focuses on content creation in various fields such as art, design, and entertainment. Let’s focus on 2 of its major use cases in finance to evaluate whether it is a blessing and/or a curse…

Do not confuse this with Artificial General Intelligence (AGI). This broader concept can be characterised by Generative AI as it replicates human intelligence and outperforms humans in various domains.

Product Development

Generative AI, particularly in the form of code generation, is being leveraged to improve internal productivity. The consensus is accuracy with code snippets yet integration issues with the broader program. Adapting the program to specific technical environments requires human programming capabilities including prompt engineering.

  • Deloitte conducted extensive experiments with GPT-3’s Codex – aimed to improve speed and effectiveness. It found that it enhances productivity for experienced developers and introduces programming capabilities to those with no experience. In a six-week pilot at Deloitte involving 55 developers, most users rated the resulting code’s accuracy at 65% or better, with most of the code being generated by Codex. This experiment revealed a 20% improvement in code development speed for relevant projects. Deloitte has also used Codex for translating code between different programming languages. They conclude that the output quality is highly dependent on the quality of the input prompts. We can learn prompt engineering to benefit from this lucrative technology.

Companies are also developing specialised tools to suit their needs:

  • JPMorgan Chase has developed two: DocLLM and SpectrumGPT. DocLLM is tailored to understand complex business documents such as forms, invoices, and reports, while SpectrumGPT analyses vast amounts of documents and proprietary research to provide insights for portfolio managers. These tools have improved document comprehension and operational efficiency, achieving a 15% performance enhancement over more ‘general’ technologies like GPT-4.

Management

  • Morgan Stanley is exploring LLM-based knowledge management in collaboration with leading commercial LLM providers. They are working with OpenAI’s GPT-3 to fine-tune training on wealth management content. This allows financial advisors to search for existing knowledge within the firm and create tailored content for clients easily. Users of such systems will likely need training or assistance in creating effective prompts, and the knowledge outputs from LLMs may require editing or review before application. Addressing these issues could significantly enhance the field of knowledge management and allow it to scale more effectively.

AI in risk management has been gaining traction too. Banks traditionally use traditional credit risk models to predict categorical, continuous or binary outcome variables. This is because ML models are difficult to interpret and are not easily verifiable for regulatory purposes. However, recently, professional judgements have been less valued than data-driven lending. Financial services firms are also increasingly hiring external consultants who use deep learning methods to develop their revenue forecasting models under stress scenarios. Trader behaviour analysis by monitoring email traffic and calendar-related data, check-in/check-out times, and call times combined with trading portfolio data, systems can predict the probability of trader misconduct, saving millions in reputational and market risk for financial institutions.

Customer relationship management is another important area for banks. They provide personalized 24/7 services to customers, by facial recognition and voice command features to log in to financial apps. Analysis of customer behavioural patterns and automatically performing customer segmentation allow for targeted marketing and improved customer experience and interaction.

  • After launching in January 2024, Klarna announced that its AI chatbot does the equivalent work of 700 full-time employees and has led to a 25% decrease in repeat inquiries as well as bringing down resolution time to under two minutes (compared with 11 minutes previously).

Overall, we understand that human oversight and expertise remain crucial. The McKinsey Global Institute (MGI) estimates that across the global banking sector, generative AI could add between $200 billion and $340 billion in value annually, or 2.8 to 4.7 per cent of total industry revenues, largely through increased productivity.

Conclusion

AI models are adaptable, flexible, and scalable but prone to bias, hallucinations, and greater complexity, which makes them less robust. They may also present false or misleading information as facts – known as “hallucinations”. None of these LLMs are perfect conversationalists. Training on past human content causes them to replicate any racist, sexist, or biased language to which they were exposed. This algorithmic bias raises serious consequences such as increased reputational or legal risks.

When deployed, generative AI can increase efficiency, but its performance is difficult to predict and subject to possible misuse or overreliance. For example, criminals could fine-tune and spoil otherwise harmless AI for specific operations including cyberattacks, misinformation, market manipulation, use of deep fakes to undermine confidence in a financial institution, etc.

Ultimately, getting generative AI right can potentially unlock tremendous value as well as complications. New regulatory initiatives are inevitably essential to tackle the market failures arising from increased endogenous risks that the current prudential framework doesn’t address. Well, it is a lot of information to process. If you do plan on working in the Financial Services industry or are just curious about AI, I recommend you keep up and gather insights through valuable sources such as AI TLDR newsletter.

Find out more:

Bibliography

Leave a Comment

Your email address will not be published. Required fields are marked *

Solverwp- WordPress Theme and Plugin

Scroll to Top