EMVC Vantage Point - Financial Services in the age of Generative AI
Highlighting the crucial role generative AI could play in financial services and the various use cases GenAI is already being used for to put to the test.
Generative AI has captured VC interest as a lucrative investment category, growing 2.8x since 2020 to about 1% of global venture investment in 2022. Generative AI using large language models (LLMs) has had a breakthrough in 2022 due to the ability to train models on vast quantities of unstructured data and ability to use large computational power to generate new insights in the context of information provided. The first phase of innovation is transforming writing, gaming, art, marketing and sales and the dev app space.
Generative AI in financial services is still a nascent technology use-case. According to one report, the global generative AI in Fintech market is currently only $850mm, expected to grow at 22% CAGR to $6.3bn by 2032. However, there is massive potential to transform multiple areas of the financial services industry, particularly as the industry operates with troves of financial data.
Financial services companies are starting to partner with generative AI foundation model companies to develop in-house solutions for multiple use cases. For example, Morgan Stanley is partnering with OpenAI to develop chatbots that can assist financial advisors. Bloomberg recently announced the development of an in-house generative AI model. This model is trained on both proprietary financial data and general data, and can assist with multiple financial tasks such as sentiment analysis and news classification. According to Bloomberg, this model outperforms general LLMs in financial-based tasks.
This development highlights the potential evolution of generative AI within the financial services industry. Since most companies leverage generative AI foundation models to build enterprise-specific solutions, there are low barriers to entry. As a result, foundation models are likely to become commoditized in the near future. Startups that can build differentiated and value-adding services on top of these models will ultimately be successful.
More instances below of fintech startups adopting LLM-based foundation models to build their own in-house solutions:
Brex, an expense management platform is partnering with OpenAI to launch a chat-based platform to generate insights and benchmarks for customers.
Stripe is also partnering with OpenAI to leverage its GPT4 model to improve customer support, fraud detection, NLP.
Lemonade the insurtech company, and PayTM are among other fintech startups that have also announced their plans to leverage generative AI to improve their current AI engine for customer servicing and fraud detection, though further details are not yet available.
Forecasting how generative AI could potentially be used in the financial services space, there are a number of use-cases where we are already seeing startup activity:
Wealth Management
Wealth management has the potential to be a key enabler to democratizing wealth management. Its ability to analyze vast amounts of data and derive market intelligence will allow it to generate unique portfolio advice for customers. It can also incorporate human context and build intuitive insights. In addition, it can also present market insights in an easy to understand way to enable customer decision-making.
Toggle is an AI-powered investment advisory platform that trains LLMs like ChatGPT on chat-based investment advisory. Toggle raised $10.5mm in Series A funding in 2021.
Portrait Analytics is another wealth management platform that offers a search platform with intuitive insights and personalized investment recommendations. It has raised a $3mm pre-seed round in Apr 2023.
Enhancing Customer Experience
One of the key ways that generative AI has been used in the financial services space is to enhance customer experience. With the help of chatbots, financial institutions can now provide 24/7 customer support without the need for human intervention. Chatbots can answer common queries, such as balance inquiries, account statements, and transaction history. Generative AI can also provide superior customer experiences through hyper personalized recommendations based on a customer's spending behavior, such as investment opportunities or savings advice. It could be used to assist a human customer service agent to deliver better customer support.
Glia offers a seamless customer interaction solution across platforms for financial service companies. It has solutions for customer onboarding, sales and personalized customer support.
Streebo uses OpenAI’s GPT model to power intuitive chatbots for enterprises across multiple industries. It offers personalized responses and suggestions to customers and also delivers insight reports to companies
Kaizan AI is a client intelligence platform that can be used across industries for client admin management. It automatically manages admin tasks while also providing client insights, highlighting potential risks in advance.
Fraud Detection
A unique use-case for generative AI is in enabling better fraud detection models. Generative AI can analyze existing datasets and create new data that derives characteristics from existing data but is not necessarily the same. This feature allows it the ability to create new datasets with new fraud patterns that can be used to train fraud AI models. There are a number of startups which create synthetic data using generative AI to assist a number of industries conduct internal analyses:
Scale: Generates synthetic data for training AI applications across industries and use-cases. Its solutions are used even by the generative AI models such as OpenAI. They have recently developed a full stack generative AI solution with Scale’s data engine to fine tune the models to enterprise-specific data.
Hazy develops synthetic data statistically similar to real data, helping companies preserve the data privacy in using data to train GenAI models. It recently raised its Series A round including banks such as Wells Fargo as investors.
Mostly AI develops synthetic data for financial service companies to train their fraud detection models and other internal AI/ML models.
FinCrime Dynamics offers multiple services to build resilience of a company’s financial crime detection solutions - it creates synthetic fraud databases and creates simulations to train fraud models.
Tonic is also a similar solution providing synthetic data to financial service companies for better internal compliance systems.
Risk management and compliance
Generative AI can help build superior risk management systems due to its ability to generate predictive analytics with more complex scenarios and data. It can also pull information from disparate sources and draw meaningful conclusions about potential risks. Due to this, startups are exploring building financial crime management solutions with generative AI models.
SymphonyAI has recently launched a platform ‘Sensa Copilot’ to accelerate financial crime detection and management. The solution gathers relevant data from disparate relevant internal sources and summarizes findings in easy to understand reports for human analysis. It can also calibrate its responses based on the organizational / situational context. This solution operates on the OpenAI model.
Underwriting models
There is also significant potential for generative AI to be used in underwriting models. By analyzing large amounts of data, generative AI can identify patterns and predict outcomes more accurately than traditional underwriting models. This could lead to more accurate assessments of risk and ultimately more informed lending decisions.
Zest Finance is a fintech company that uses generative AI to develop dynamic and nuanced underwriting models. It allows building custom credit models and enables nuanced risk prediction during the underwriting process. It uses a generative adversarial network (GAN) model.
Conclusion
The financial services industry is likely to adopt generative AI incorporated within solutions that improve the efficiency and effectiveness of operations, rather than directly adopting generative AI models. This is due to the higher requirements for data accuracy and risk aversion. Bain Capital Ventures has identified two types of generative AI applications: Generative AI as a Service (GaaS) products, which are built on top of foundational models with general use-cases, and Generative AI as a Component (GaaC), which are built for specialized purposes and have a strong product/workflow built on top of the basic app functionalities. Financial services companies are more likely to adopt GaaC solutions, as they can use generative AI as a component within their existing systems to improve operational performance while maintaining control over the accuracy and quality of output.
In this context, it is interesting to see how the first wave of startups are attempting to create a product and workflow around widely available foundational models to improve specific industry functions. Startups are still exploring different types of component-based solutions and it will be interesting to see which app areas continue to grow.
By Akshara Anand, Venture Fellow at Emphasis Ventures and student at Columbia Business School class of '24.