Generative AI in Financial Services

most popularAI

May 3

In the last few months, there’s an increased buzz from Generative AI (artificial intelligence) across all technology companies.

The modern financial services industry has leveraged AI and machine learning since early 2010s as a way of improving fraud management, credit underwiring, and user onboarding.

Using large language models (LLM) like ChatGPT, Gen AI is proactively creating new content instead of filtering through existing data to reactively make a prediction. The potential is starting to be seen with Chat-GPT, but there’s specific areas and use cases in fintech that would make this a gamechanger. Here’s a brief.

PREMISE FOR GEN AI in fintech

What many outside of AI don’t realize is that the potential change would be more of a transformation for banking services. The combination of large volumes of raw data (most of it not being utilized) with massive computing capabilities will reach further farther than innovation from mobile access. Both new startups and established companies are expected to adopt some form of Gen AI this decade.

With the volume of untapped financial data available at financial institutions, there’s a strong fit for new models to be created and trained which are capable of responding to most product inquiries (user experience), and identifying suspicious financial activity (compliance). Within these two area, there are opportunities to go further and completely personalize user experiences with custom recommendations & advice, boost existing risk management screening & compliance policies, and improve forecasting in real-time.

Well-established banks and financial institutions already have access to more financial data than newly formed startups. Companies less than 5 years old would need to rely on public data in starting to form their own models.

USE CASES

Personalized experiences is usually front & center when we think about FinTech and the advances made with interfaces and apps. However, there’s still a lag in terms of specific customization that impacts a user and their financial life. This is attributed to the need to contextualize transaction data and interpret patterns into a strong recommendation that compares benefits and trade-offs. A black and white decision isn’t always available when it comes to finances — being able to navigate through the gray is what’s missing.

Gen AI solutions are being developed to help customers maneuver multiple, smaller financial decisions towards a larger goal. By being able to take multiple questions, make comparisons, and add human context, we should see a leap forward in advising for the everyday consumer.

Optimizing business operations is also being addressed by Gen AI, which is able to pull data from various sources and interpret it through a lens of compliance laws and personal financial decisions. The difficulties in the past came from automating complex choices that are often emotional and ensuring compliance with regulation across all employees.

Banks staff large customer service and support teams to respond to client inquiries which may involve product information and compliance guidelines. New staff can access robust resources in a trained model built off of 8+ years of actual customer calls. A repository for responses helps build knowledge and confidence quickly.

Mortgage and business loan divisions must also search various sources to build a file for applicants and submit for approval. Instead of spending time sifting through databases, loan officers can make the inquiry through an LLM model and focus on ensuring accuracy of the file. Faster and efficient loan processing increases capacity for teams. By training models with guidance from multiple regulators, potential issues can also be quickly found and remedied before a file is processed.

Broadening the discussion of compliance, generative AI can make a direct impact in minimizing illicit financial activity and money laundering. Companies already invest a significant amount in programs that leverage rules-based systems to flag transactions and users. Many of these potential matches require manual review from compliance employees and lead to false positives. Manual processing requires requesting or accessing more data to validate (or invalidate) a match. Poor or inadequate compliance reviews can lead to fines and penalties for financial institutions and fintechs.

Gen AI in compliance fields can help:

Improve due diligence reviews: Large volumes of documentation (statements, activity reports, filings, emails, contracts) can be analyzed continuously to ensure ongoing adherence to regulation and reduce the need for deeper reviews;

Data access: Compliance officers often struggle to string together fragmented data from multiple sources, which is needed to make a final decision on a flagged transaction or customer. Improved LLMs can quickly reduce the time needed for research and response;
Enhance money laundering detection: Training models on past reports of submitted of suspicious activity to uncover new patterns;

Overall, risk management is a critical area in the current market environment. Financial institutions and fintechs need help with measuring & monitoring for various types of risk (operational, regulatory, credit, market, data security, etc.). Support in finding areas of risk and quickly responding will improve drive efficiency for all companies in the near term.

Capturing and processing high volumes of data from miscellaneous sources (news, industry reports, case studies, quarterly earnings analysis, etc.) through custom LLMs provides a rich view of multiple areas for risk managers.

This view can lead to better insights that take into account economic & market conditions, political movement, and consumer sentiment on a real-time basis. Being able to change certain variables towards a more conservative climate would yield a predictive forecast for risk teams — accounting for further increases in inflation across specific countries.

These four areas (personalized experiences, business operations, compliance, risk management) stand to gain the most from generative AI integration in financial services. There are still challenges for early adoption.

progress in adopting gen AI?

ChatGPT has been a great tool to showcase what’s possible with LLMs and generative AI in the last year. Many have signed up and tried out complex queries, which yielded comprehensive responses quickly — saving time in researching online and sifting though numerous articles.

The initial challenge for adoption is connected to financial data. Financial institutions have years of unstructured data from actual customers and transactions, which are critical for narrowing models to specific use cases. This offers a headstart in building strong LLMs, but these larger players conservatively approach innovative shifts — evaluating a new project, gaining approval, allocating resources, integration, and testing can take years.

Startups are built to act quickly and leverage innovation to build a successful platform for new customers. Despite being new to market and having no existing customers to provide data, online sources with public info will be utilized in creating new models. Over time, actual customer data can be plugged in to customize LLMs.

Once models are built and producing results, the next challenge is accuracy. After churning through data for a response, is the LLM providing a relevant, coherent answer? In banking and financial recommendations, there’s a higher threshold for accurate answers. Customers may be asking about tax and investments in which there can be expensive consequences in using wrong information. For early adoption, some human supervision and verification will be needed.

Overall, consumers & businesses won’t feel the immediate impact from Gen AI to their everyday lives. Financial apps and banks will improve recommendations and insights slowly, which will appear as notifications. Actual financial plans with decisions driven by AI is still a few years away, but likely the biggest gamechanger for customers globally — especially those who don’t qualify for a dedicated (human) financial advisor.