Today, with the rapid development of technology, AI (artificial intelligence) and ML (machine learning) have been galloping in the financial services industry for more than a decade, covering various improvements from better risk control to basic anti-fraud scoring. Today, generative AI based on Large Language Models (LLMs) represents a historic leap that is changing education, games, business, and many other fields. Unlike traditional AI/ML, which mainly makes predictions or classifications based on existing data, generative AI is able to create entirely new content.

Imagine being able to train on massive amounts of unstructured data, coupled with virtually unlimited computing power, which could bring about the biggest change in the financial services market in decades. Unlike in other platform shifts—such as the Internet, mobile, cloud computing—the financial services industry is always one step behind in adoption, and here we expect to see the best new and existing companies embracing generative Sex AI.

Financial services firms have vast amounts of historical financial data, and if they use this data to fine-tune LLMs (or train them from scratch like BloombergGPT does), they will be able to quickly answer almost any financial question. For example, an LLM trained on a company's customer chats and some additional product specification data should be able to answer all questions about that company's products in a split second, while an LLM trained on a decade of company suspicious activity reports (SARs) The LLM, should be able to identify a series of transactions that may signal a money laundering scheme. We believe the financial services industry is poised to leverage generative AI to achieve five goals: personalized consumer experiences, efficient operations, better compliance, improved risk management, and dynamic forecasting and reporting .

In the race between incumbents and start-ups, when AI is used to launch new products and improve operations, incumbents have an initial advantage because of access to proprietary financial data, but they will eventually lose because of High standards of accuracy and privacy are hindered. New entrants, on the other hand, may initially need to use publicly available financial data to train their models, but they will soon start generating their own data as an icebreaker for new product distribution.

So let’s dive into these five goals and see how incumbents and startups are leveraging generative AI.

Personalized Consumer Experience

Despite their enormous success over the past decade, consumer fintech companies have yet to deliver on their most ambitious promise: optimizing consumers’ balance sheets and income sheets without human involvement. This promise has not been fulfilled because user interfaces fail to adequately capture the human context that influences financial decisions, nor provide advice and cross-sell in a way that helps people make the right trade-offs.

An important example of one of these non-obvious human situations is how consumers prioritize paying their bills during difficult times. Consumers typically consider both utility and brand when making this decision, and the intertwining of these two factors complicates creating an experience that adequately captures how to optimize that decision. This makes it difficult to provide top-notch credit education without the involvement of human employees. While an experience like Credit Karma can take the customer 80% of the journey, the remaining 20% can feel like a magical abyss, and further attempts to capture the context tend to be too narrow or use false precision, thereby undermining the consumer trust.

Similar deficiencies exist in modern wealth management and tax preparation. When it comes to wealth management, human advisors trump fintech solutions, even those narrowly focused on specific asset classes and strategies, because people are deeply shaped by their unique hopes, dreams and fears. This is why human advisors have historically been better at tailoring advice to their clients than most fintech systems. When it comes to taxes, even with the help of modern software, Americans still spend more than 6 billion hours a year processing their taxes, make 12 million mistakes, and often omit income or forgo benefits they don't know about, such as possible deductions Work travel expenses .

Large language models (LLMs) offer a neat solution to these problems by better understanding and navigating consumers' financial decisions. These systems can answer questions (“Why is some of my portfolio in municipal bonds?”), evaluate trade-offs (“How should I think about term risk versus return?”), and ultimately incorporate human context into decision-making (“Your Can you create a plan that is flexible enough to help me financially support my aging parents at some point in the future?"). These capabilities should transform consumer fintech from a high-value but narrow set of use cases to another that helps consumers optimize their entire financial lives.

efficient operation

In a world where generative AI tools can infiltrate banks, Sally should be continuously evaluated so that she has a pre-approved mortgage once she decides to buy a house.

However, this world has not yet materialized, for three main reasons:

• First, consumer information is scattered across different databases . This makes cross-selling and forecasting consumer demand extremely challenging.
• Second, financial services are highly emotional purchases that often have complex decision trees that are difficult to automate . This means that banks have to employ large customer service teams to answer the many questions their customers have about which financial products are best for them based on their individual circumstances.
• Finally, financial services are heavily regulated . This means that human employees like loan officers and processors have to be in the cycle of every available product (like a mortgage) to ensure compliance with complex and unstructured laws.

Generative AI will make labor-intensive functions such as extracting data from multiple places and understanding unstructured personal context and unstructured compliance regulations 1,000 times more efficient. for example:

• Customer Service Representatives : At every bank, thousands of customer service representatives must have detailed knowledge of the bank's products and related compliance requirements in order to be able to answer customer questions. Now imagine a new customer service representative starts work, and they have access to a large language model (LLM) trained on the past 10 years of bank customer service calls. The rep can use the model to quickly generate the correct answer to any question, helping them talk more deeply about a wider range of products while reducing their training time. Existing firms will want to ensure that their proprietary data and personal information of specific customers is not being used to advance a generic LLM that other firms can use. New entrants will need to be creative in how they structure their datasets.
• Loan Officers : Existing loan officers typically need to pull data from close to a dozen different systems to generate a single loan document. A generative AI model could be trained on data from all these systems, so that a loan officer could simply provide a customer name and loan documents would be instantly generated for them. Loan officers may still need to ensure 100% accuracy, but their data collection process will become more efficient and accurate.
• Quality Assurance : Much of the quality assurance work for banks and fintech companies involves ensuring full compliance with numerous regulators. Generative AI can greatly speed up this process. For example, Vesta can use a generative AI model trained on Fannie Mae’s sales guide to immediately alert mortgage loan officers of compliance issues. Since many regulatory guidelines are publicly available, this may provide an interesting entry point for new entrants to the market. However, the real value will still flow to those companies that own the workflow engine.

All of these are steps toward a world where Sally has instant access to potential mortgages.

better compliance

In the future, if the compliance department can adopt and use generative AI technology, the global annual illegal money laundering of 800 billion to 2 trillion US dollars may be effectively prevented. Drug trafficking, organized crime and a variety of other illicit activities could see the biggest declines in recent decades.

Today, we spend tens of billions of dollars a year on compliance, but actually prevent only 3% of criminal money laundering. Most compliance software is built based on "hard-coded" rules . For example, anti-money laundering systems allow compliance officers to enforce rules like "flag any transaction over $10,000" or look for other preset suspicious activity. But applying these rules is often not ideal, because many financial institutions are legally required to investigate a large number of false positive situations, which are often complex and difficult to deal with. In order to avoid heavy fines, the compliance department employs tens of thousands of employees, usually accounting for more than 10% of the bank's total staff.

And once we can take advantage of generative AI, the future scenario will change:

• More efficient screening : The generative AI model can quickly aggregate the key information of any individual in various systems and present it to compliance officers, enabling them to conduct risk assessments on transactions more quickly.
• Better predicting money launderers : Imagine a model trained on Suspicious Activity Reports (SARs) from the past 10 years. Without explicit instructions, the AI can discover new patterns from the reports and define for itself what behavioral patterns are likely It is money laundering.
• Faster document analysis : The compliance department is responsible for ensuring compliance with the company's internal policies and procedures, as well as meeting regulatory requirements. Generative AI can analyze large amounts of documents, such as contracts, reports, emails, etc., and then flag possible problems or areas that need further research.
• Training and education : Generative AI can also be used to develop training materials that simulate real-world scenarios to teach compliance officers how to conduct best practices and how to identify potential risks and non-compliance.

New entrants can leverage publicly available compliance data from dozens of agencies, making searches and integrations quicker and easier. And for those large companies with years of data accumulation, they need to design appropriate privacy features.

Improved Risk Management

While the Archegos and the London Whale sound like creatures from Greek mythology, they actually represented a critical failure of risk management that cost some of the world's largest banks billions of dollars. Coupled with the recent example of Silicon Valley Bank, it becomes clear that risk management remains a major challenge for many leading financial institutions.

While advances in AI cannot completely eliminate credit, market, Liquidity, and operational risks, we believe this technology can play an important role in helping financial institutions identify, plan for, and respond to these inevitable risks more quickly. Specifically, here are some areas where we believe AI can help with more effective risk management:

• Natural Language Processing : LLM models like ChatGPT can help process large volumes of unstructured data such as news articles, market reports, and analyst research to provide a more complete view of market and counterparty risk.
• Real-time insights : Instant understanding of market conditions, geopolitical events and other risk factors allows companies to adapt more quickly to changing conditions.
• Predictive analytics : The ability to run more complex scenarios and provide early warnings can help companies manage risk more proactively.
• Integration : Bringing disparate systems together and using AI to consolidate information can help provide a more complete view of risk exposures and streamline the risk management process.

Dynamic Forecasting and Reporting

In addition to helping solve financial problems, LLM can also help financial service teams improve their internal operating processes and simplify the daily work steps of the financial team. Even with all the other aspects of finance making strides, modern finance teams still rely on Excel, email, and human-intensive business intelligence tools for their day-to-day workflows. The automation of basic tasks is hampered by a shortage of data science resources, and CFOs and their teams are overwhelmed with tedious record-keeping and reporting tasks, instead of focusing on more important top-level strategic decisions.

In general, generative AI can help these teams pull data from more sources and automate the process of highlighting trends and generating forecasts and reports. The following are some specific application examples:

• Prediction : Generative AI can help write formulas and queries in Excel, SQL and BI tools to automate analytics. In addition, these tools can help uncover patterns, distill predictors from larger datasets with more complex scenarios, such as macroeconomic factors, and suggest how these models can be more easily tuned to inform corporate decision-making.
• Reporting : No need to manually extract information from various types of data for reporting (such as board reports, investor reports, weekly data panels), generative AI can help automatically create text, charts, graphics, etc., and flexibly adjust report content according to different examples .
• Accounting and Taxation : Accounting and taxation teams need to spend a lot of time referencing regulations and understanding how they apply to real situations. Generative AI can help synthesize, generalize, and suggest possible answers to tax laws and potential tax cuts.
• Purchasing and Payables : Generative AI can help automate the generation and adjustment of contracts, purchase orders and invoices, reminders, and more.

However, we need to be clear that current generative AI has its limitations in areas that require judgment or precise answers (which is often a must for finance teams). Generative AI models continue to advance in computing power, but we cannot yet fully rely on their accuracy, or at least require human review. With the rapid improvement of models, more training data and the ability to combine mathematical modules, new usage possibilities are presented.

challenge

Within these five megatrends, new entrants and existing market players face two major challenges to realize this generative AI-based vision of the future.

• Training Large Language Models (LLMs) to handle financial data : LLMs today are mostly trained on web data. These models need to be fine-tuned with financial data to meet the specific needs of financial services. New entrants may start with publicly available corporate financial data, regulatory filings, and other readily available public financial data, further refine their models, and then gradually use their own collected data over time. Incumbent players like banks, or large platforms with financial services businesses (such as Lyft), who can leverage the proprietary data they already have, may give them some initial advantage. However, incumbent financial services firms tend to be too conservative in embracing large platform changes. This provides a competitive advantage for unfettered new entrants.
• Accuracy of model output : These new AI models must be as accurate as possible, given the impact that answers to financial questions can have on individuals, companies, and even society as a whole. They can't make up wrong answers, or give answers that sound confident but are wrong, and they need to be more accurate than pop culture queries or the average high school essay for critical questions about people's tax or financial health. precise. Initially, there is often a need to have a human in the loop as the final validation of the AI-generated answers.

The rise of generative AI is undoubtedly a major platform change for financial services companies. It has the potential to provide customers with more personalized solutions, make company operations more cost-effective, improve compliance, improve risk management, and at the same time It can also lead to more flexible forecasting and reporting. Incumbents and startups will compete on the two key challenges we just listed. While we don't yet know who will win out, we already know that one clear winner has emerged: the consumers of financial services of the future .