Why financial services?
Financial services’ unique mix of data-driven decision-making, high-volume transactions, risk management requirements, and complex regulatory landscapes make it a compelling case study for the application of these advanced AI systems. Additionally, over the last few years, we’ve seen a sharp increase in the adoption of financial digital solutions for both consumers and businesses, as well as an explosion of availability of financial-related data. The emergence of open banking, spearheaded by platforms like Plaid and Finicity, laid the foundation for this data revolution. We are now witnessing a further evolution of this in the “Plaid for X” model, where companies like Finch, Codat, Rutter, Merge, Pinwheel, and Argyle are enabling open data access for various financial-related sources beyond traditional banking.
We believe this abundance of financial-related data, as well as the financial services industry’s increasing need for speed and personalization, will create a new generation of AI-first fintech companies that will significantly change the way the entire industry works.
However, not all subsets of the financial services industry will benefit from and embrace AI at the same rate. In the following section, we will share the framework we have developed internally to guide our thinking in this fast-changing environment. We hope you’ll find it helpful.
Our framework: Evaluating opportunities for AI in financial services
As a first step, we broke out the financial services industry into major categories (Financial Management, Lending, Insurance, Payments, Trading and Wealth Management, and Risk). For each, we selected the most important functions within each category (for example, within Insurance, we identified Insurance Underwriting, Distribution, and Claims & Customer Service). We dove deep into each function, talking to users to understand their pain points, and imagined how each “job to be done” will change with AI.
After examining all functions, we ranked them based on the following three criteria:
- Incremental AI Unlock. This category is meant to assess the extent to which AI can enhance existing solutions, processes, or business models. Using a first-principles approach, we considered how AI technology's ability to read or write text-based data would impact key functions, as well as its ability to find patterns, categorize, and understand large volumes of data. When doing this, we considered all aspects of the “job to be done” that could benefit from the potential AI unlock. How could AI significantly enhance the experience for the end customer? Going back to the insurance underwriting example, could an AI-powered underwriting solution help a CEO more efficiently purchase a cyber insurance policy for their company? Or, how could AI improve the experience of the user within an insurance company (in this case, the underwriter of the cyber insurance policy?) The more potential incremental AI unlock, the more enterprise value could be created, making this an attractive function for entrepreneurs to focus on.
- Urgency of Customer Pain.The more urgent the customer pain is, the more interesting the category is. While the previous criteria were focused on the technological upside that AI can provide, here we assessed the reality of the market and demand for a given solution. Existing strong and pressing customer pain is typically associated with a higher willingness to pay and shorter sales cycles, both desirable elements of a market for an entrepreneur. We also considered whether there are any trends, supported by regulatory or technological shifts, that will change the urgency of the customer pain in the near term. For example, as AI technology advances, we foresee significant challenges in the realm of fraud prevention due to its potential misuse. We expect unscrupulous actors to leverage AI for nefarious uses, escalating the complexity and prevalence of fraudulent activities and therefore requiring more sophisticated solutions to combat it. Another example of a supporting trend, this time in payments, is FedNow, an initiative by the Federal Reserve to build a real-time payments system designed to facilitate instant monetary transactions on a 24/7 basis. As financial institutions participate in this initiative, they will have to rethink some of their tech stack to be able to process transactions instantaneously.
- Function Risk Tolerance. The last criteria we considered was a function risk tolerance. Said in a different way, how high are the stakes for each function? To determine this, we considered what the implications and outcomes would be if AI were to incorrectly perform the function in our framework. For example, if the anti-money laundering (AML) function within a bank was to be incorrectly performed by AI, even just once, the implications could be catastrophic for the financial institution, as they could lose their license or be forced to pay exorbitant fines. Therefore, AML is a function with an inherently low risk tolerance. On the other hand, if AI were to incorrectly personalize a financial services offer or incorrectly answer a customer service inquiry, it would be damaging to the brand and the customer, but the risk to the company would not be existential. Therefore, distribution or customer support are functions with higher risk tolerance.
We believe that everything else being equal, functions with a higher risk tolerance will likely see faster adoption of next-gen AI-powered solutions, while for lower-risk ones, buyers will want to see more proof points around models' performance and accuracy. It’s worth noting that there can be some tension between this category and the previous one: even if a function has low risk tolerance, it can still be a highly desirable place to build a company because of the massive pain points waiting to be solved. This is the case with functions such as KYC, AML, and compliance, amongst others.
Based on this framework, we identified a few areas that we think present particularly attractive near-term opportunities for founders looking to build AI-first companies:
Fraud (Onboarding, KYC/KYB, AML, fraud monitoring, and compliance)
- Risk teams today often have to parse through large sets of text and unstructured data. Through AI’s improved reading and writing abilities, risk teams can make more accurate decisions and generate quicker reports (like an AML report).
- Furthermore, Generative AI has enhanced fraudsters’ social engineering abilities, so a new generation of tools focused on detecting AI-enhanced fraud needs to be built.
- Given the strong existing (yet growing) pain around fraud in financial services, there has already been a lot of startup activity in the space with fraud solutions and risk orchestration layers like Sardine, Alloy, Oscilar, Sandbar, and more.
- Because AI can more accurately classify transactions and entity data, there are opportunities to help accounting teams with transaction categorization, anomaly detection, and reporting (like flux explanations).
- Some notable recent launches of AI in accounting are Ramp’s recent Ramp Intelligence product suite, Numeric, and Basis.
Insurance is an industry that is inherently data-driven and predicated on the ability to assess risks accurately, making it an excellent fit for AI-powered solutions. Below are a few examples.
- Underwriting: By collating and analyzing a myriad of data points, AI can optimize the underwriting process, streamlining data interpretation and enhancing decision-making efficiency. In this context, startups like Federato are pioneering the use of AI to align underwriting decisions with broader corporate objectives, while others like Taktile are building infrastructure to support ML-powered underwriting.
- Claims: AI can drastically improve the claims handling process, making it faster, easier, and more objective. Companies like Hi Marley, Tractable, and EvolutionIQ are utilizing AI and machine learning to expedite claims processing, reduce human error, and detect fraudulent claims. This leads to increased customer satisfaction and reduced operational costs.
- Distribution: AI enables the hyper-personalization of insurance products, allowing for tailored policies that cater to individual needs and circumstances. This could particularly revolutionize lines of insurance that have traditionally been challenging to place, such as personal life insurance. By leveraging AI, insurers can optimize product offerings, enhance customer engagement, and ultimately improve business outcomes.
- LLMs’ improved ability to discern meaning and relationships behind text (like transaction descriptions) can significantly improve payment risk to reduce chargebacks, fraudulent transactions, and insufficient funds risks. There have already been success stories from early movers such as Slope, who shared their impressive results from incorporating LLMs into their internal payment risk models.
- Some other notable startups that have applied AI to improve payment risk models are Sardine, Spade, and Coris.ai.
- LLMs will have a tremendous impact on financial research thanks to their ability to parse, analyze and draw insights from vast amounts of data accurately and rapidly. By automating the extraction of key data points from publicly available sources—such as financial reports, news articles, and market updates—they can greatly accelerate the research process, reducing the time spent on data gathering and enabling financial analysts to focus more on strategic decision-making. Furthermore, LLMs can assist in identifying trends, patterns, and correlations that human researchers might miss, adding a new layer of sophistication to financial analysis.
- Next-gen, AI-first companies in this space include Hebbia, Brightwave, Brox, Fintool, Quill, Portrait Analytics, Alphawatch, and Stratosphere, amongst others.
As we think about how the space is evolving, there are a few open questions that remain:
First, what financial services-specific infrastructure is needed to enable more development and adoption of these tools? Will there be even more demand for financial-related data providers like Plaid, Codat, or Finch to help build better models? How will organizations think about training models on internal versus external data?
Second, as AI systems become more deeply ingrained in our financial lives, establishing and maintaining trust will be paramount. How will these AI solutions demonstrate their reliability, accuracy, and transparency over time to inspire confidence among both buyers and customers?
Furthermore, will they be able to walk the line between personalization and privacy, a delicate balance in today's digital age?
Lastly, there are still many unknowns in regard to the regulatory environment. Will the current and future regulatory environment permit the widespread use of AI in the financial sector? What limitations might apply? And how will regulators ensure that these AI applications are equitable, preventing discriminatory practices and ensuring fairness in an increasingly automated financial world?
We are excited about the potential AI holds in reshaping the financial services industry, empowering companies with unprecedented efficiencies, and enhancing the user experience across the board. As the landscape continues to evolve at a rapid pace, we will continue to monitor new changes and how they may affect the financial services industry.
Are you a founder building in this space? Or are you considering doing so? We’d love to hear from you! Feel free to reach out to us at email@example.com and firstname.lastname@example.org and we’d be delighted to chat.
Enjoying this article?
Sign up to gain access to our thought leadership and have future articles delivered directly to your email.