Generative AI in fintech: Do the benefits outweigh the risks?

Potential:

How might generative AI play out in fintech? Financial services is a highly regulated field, and we think the impact of generative AI will, in the short term, be limited largely to providing better customer service and faster onboarding. In the longer term, we believe the potential transformational impact of generative AI should be seen in the context of a much wider trend in fintech: lower transactional friction.

In 1923, in his seminal paper “The Nature of the Firm,” economist Ronald Coase discussed the importance of “transaction costs” to the shape and nature of economies. The greater the costs associated with negotiating, enforcing, and overseeing a job — that is, the more friction associated with a transaction — the more likely a firm is to either do that job internally or skip doing it altogether. Economies with a lot of friction tend to be centered on large, vertically structured firms, tend to be built on mechanisms of informal trust like family, and, in general, tend to be less innovative. Conversely, economies with lower friction tend to favor smaller, more nimble firms that depend on large supplier networks, and, in general, tend to be more innovative. For this insight, Coase won the Nobel Prize in 1991.

In fintech, transaction friction has been rapidly reducing due to three major trends: the acceleration of digital financial products, open banking, and real-time payment adoption. The amalgamation of those trends may be the catalyst for unlocking the transformative potential promised by generative AI for financial services.

The Global Pandemic played an important role in accelerating the digitalization of financial services as more of the economy moved online. Recently, the adoption of other technology tools, such as online notarization with DocuSign or meetings with Zoom, has eliminated many of the physical requirements to opening a financial account, buying a home, or completing legal documents. This greater acceptance of online products has opened the door for a new generation of tools to further automate aspects of one’s financial life, a trend that could accelerate with AI. For example, moving money between accounts to maximize interest payments, optimize investment accounts for tax purposes, and flag financial fraud. Additionally, removing the friction to attain financial advice will be particularly beneficial to those who lack access today, especially elderly and rural populations who tend to have specific financial needs.

The second major trend, Open Banking, increases personalization in financial products by expanding the accessibility of financial data. Begun in earnest with Europe’s PSD2 open banking initiatives continue to gain traction globally, giving consumers more control of personal financial data and encouraging financial institutions to innovate. As a result, financial institutions around the world have been required to create APIs so that fintech start-ups can build on top of their data. The proliferation of open-banking regulation has resulted in extensive investment in the underlying infrastructure of the fintech ecosystem. This is critical for Generative AI use cases, especially in consumer applications, because it facilitates the exchange of permissioned data that can be used to design individualized offerings. In many cases traditional credit requirements, like W-2s, fall short of capturing the financial picture of gig economy workers and we see a significant opportunity to pair open-banking infrastructure with generative AI to service this gap.

Lastly, proliferation of government-backed real-time payment networks is reducing the friction of moving money, both in terms of speed and cost. Early success in Brazil (PIX), India (UPI), and other emerging economies has accelerated innovation in account-to-account payments domestically, reducing the need for traditional financial institutions to store and transfer money. While the trend will take a few years to ramp up, we believe it will become a new normal and that eventually international links between networks, already forming regionally, will allow money to move anywhere at a much lower cost in near real time. Freeing payments and radically reducing interchange fees are essential to liberating money and allowing AI to operate at a scale and speed currently unimaginable. People in emerging markets or with lower savings will particularly benefit from fast, efficient payment networks.

If generative AI is to live up to its potential, it cannot be limited to offering marginally better products to wealthy, relatively well-served, sophisticated consumers. Over half the world has been effectively left out of the modern financial system — either underbanked or unbanked — and many more have been mis-serviced or exploited by it. Building on increasingly lower financial friction to offer these groups access to new and superior financial services is fintech’s greatest modern opportunity to unleash human potential. Providing automated, personalized digital financial offerings quickly and inexpensively is perhaps the best way for generative AI to live up to its hype and change the financial world. Fintech startups, especially those that power SMBs and the underbanked, are well positioned to lead this transformation.

Opportunity:

Unlocking services (Digital Financial Services)

Over 50% of younger consumers want Generative AI to help them manage their finances, according to a new report published by Marqeta. The promise of Generative AI is its ability to understand customer preferences, spending habits, financial goals, and provide personalized financial solutions or recommendations to any individual. Given this framework, we see three areas of potential opportunity:

· WealthTech: As wealth management products become commoditized, we expect the demand for better customer service and tailored advice to accelerate. Generative AI can be used to differentiate service, expand financial education, and bolster corporate training using financial data in the near term. It will also bolster the productivity of existing teams and increase the earnings potential of individual advisors.

· CFO Stack: In the current economic climate, CFOs are seeking profitability through better cost optimization and more sustainable balance sheet management. However, these middle and back-office finance teams continue to rely on legacy systems and disjointed workflows. Generative AI can play a role in increasing data interoperability, automation, and user experience so employees can operate more efficiently. We also expect this technology to be applied outside of cost-cutting measures as revenue generating teams utilize Generative AI to better understand customers, engage with key partners, and optimize engagement.

· Onboarding/Compliance: Generative AI will also improve data capture and analysis at customer onboarding and compliance checks.. The technology can be used to ingest data from large volumes of documents, such as contracts, reports, and emails. Better data and models will enable fintechs to minimize exposure to risk while reducing negative customer experiences through false positives. We expect these applications to be limited to data entry and onboarding automation in the very near term as financial companies have low tolerance for error for KYC/AML monitoring and high reputational risk.

Unlocking connectivity (Open Banking)

The first set of applications built on open-banking rails aimed to utilize consumer data to expand credit boxes and better underwrite consumer risk. For example, fintechs aimed to use cash flow data to better underwrite repayment risk for consumers with thin or no credit history. Instead of relying exclusively on a FICO score, the consumers cash flow data could be analyzed with a comprehensive view of cash inflows and outflows both historically and in real time.

In theory, this concept makes a lot of sense. However, in practice it is very challenging to clean, structure, and analyze the underlying financial data. Because generative AI recognizes patterns in language and adapts based on the inputs of its models, many of these tasks can be made more scalable.

For example, Temenos’ latest Generative AI solution allows banks to automatically classify and label customer transactions from free-text descriptions accurately, even in various languages. These types of applications will expand with the adoption of open-banking infrastructure and drive efficiencies at scale.

Unlocking transactions (Real-Time Payments)

Many generative AI innovations within the payments space promise to significantly reduce friction in the money movement process, providing users with seamless and secure experiences. Several emerging payment trends that rely on digital technology are ideally situated to take advantage of these AI improvements. These trends include digital wallets, mobile payments, contactless and biometric payments, as well as other advanced solutions for transferring and safeguarding value. By doing so, they have the potential to usher in a new era of immediate and convenient payment choices.

For example, Visa has introduced RTP Prevent, a system that employs AI to examine transaction data in real-time and swiftly assess potential risks for the financial institutions handling real-time payments. With enhanced risk analytics, financial institutions decide whether to approve transactions with better information, which, in turn, helps in detecting and preventing fraud before it occurs. As a result, it enhances the overall safety and security of real-time payment networks.

Risks:

A recent report by the International Monetary Fund (IMF) on “Generative Artificial Intelligence in Finance: Risk Considerations” urges financial regulators to enhance their institutional capabilities and their oversight and monitoring of the development of generative AI. The report focuses on five risks:

· Embedded bias — Historical biases can be replicated in training datasets or codified via algorithm design decisions. Such bias might become ingrained in user-entered prompts for generative AI applications, with the potential for this input data to shape the generated outputs. Unlike AI systems used for predictive purposes, generative AI relies on user prompts to craft responses based on probability.

· Hallucinations — Generative AI has the capacity to “produce wrong but plausible-sounding answers or output and then defend those responses confidently.” Whether generative AI is employed for producing risk assessments, customer profiling, gathering market insights, or any other application, inaccuracies or “hallucinations” in its outputs could substantially jeopardize a financial institution’s stability and reduce the level of consumer protection it provides to clients.

· Use of synthetic data — While synthetic data can be useful for increasing data privacy and efficiently acquiring new data sets, the utilization of synthetic data comes with inherent risks. Notably these datasets can replicate real-world biases, data gaps, or large-scale inaccuracies.

· Explainability — Generative AI is “exacerbating the explainability problem.” Not all generated output from generative AI can be accurately mapped to granular decisions of algorithmic design or training data choices, making it hard to ensure “the explainability of decisions and actions taken as an outcome of AI algorithms.” This is especially problematic in financial services where model risk management procedures require explainability.

· Data privacy — Depending on large language models (LLMs), generative AI systems have the capacity to draw inferences even when the training dataset is no longer available or has been discarded in the context of system usage. As a result, information about individuals from the training dataset may be “retained” even after the data has been utilized and discarded.

Conclusion:

Despite the IMF’s caution regarding inherent risks and the potential negative effects on the financial sector, it is likely, absent government intervention, that over time AI-powered fintech solutions will gain widespread adoption within the industry. Financial service firms are already studying the adoption of AI and machine learning to capitalize on the data from new digitally driven channels. In fact, an Economist Intelligence Unit research report found that 86% of financial services executives plan on increasing their AI-related investments through 2025.

· The survey identifies investment banks as pioneers in AI adoption, with retail banks closely following suit. Insurance adoption has lagged, possibly due to slower adoption of cloud-based technologies compared to its banking peers. .

· Globally, 37% of financial services firms utilize AI primarily to cut operational costs. Additionally, there is a strong focus on employing AI for predictive analytics to enhance decision-making and increasing employee capacity to manage volume-intensive tasks.

· Financial services firms are currently constrained in their broader adoption of AI technology primarily due to its high costs, including regulatory hurdles, followed closely by inadequate infrastructure and data quality.

Opportunity maps in the intersection of generative AI and fintech:

LLMs for fintech

While companies such as OpenAI, Google, Anthropic, and others are likely to take significant strides in the development of general knowledge graphs in AI, an equally intriguing trend sees formidable players concentrating on specific domains and application scenarios. These entities may still be developing LLMs, but their primary emphasis is on achieving depth rather than breadth in their scope.

The most compelling rationale for adopting more specialized approaches lies in their potential to yield superior results tailored to a company’s precise requirements. Furthermore, such approaches may prove to be more cost-effective to operate, as their smaller language model parameters demand less computational power.

AI has disrupted various industries, but the amount of unstructured financial data creates many use cases that need finance-specific, generative AI. Companies such as Cognaize have the capability to efficiently analyze extensive volumes of unstructured financial data. They can extract insights with remarkable precision and speed, leading to improved decision-making, enhanced risk assessment, and the discovery of patterns and trends that were previously hidden due to complexity and human errors. A VC investor in Cognaize summarized the opportunity the company is pursuing by saying that “AI has disrupted various industries, but the massive amount of unstructured financial data creates countless use cases that need finance-specific generative AI”

Some startupss are taking more direct approaches, building mission-centric fintech startups powered by AI. For example, Kodex AI developed a solution that addresses the specific needs of a highly regulated industry and has the potential to significantly enhance the efficiency of how financial data is being extracted and analyzed. Its vision is to develop a copilot tailored to the specific needs of financial services professionals, enabling them to find the right information and make decisions faster.

While fintech-driven LLMs are still in their early stages of development, we see a significant opportunity for domain-specific security solutions for financial LLMs. This opportunity arises because many enterprises have implemented policies prohibiting their employees from uploading any proprietary data.

AI-powered fintech applications for better customer experience

When a LLM is trained with domain-specific knowledge, it creates the opportunity for a wider range of applications, including document automation, investment research, robo-claims processing, and wealth advisory services. Recently, we have observed significant untapped opportunities in areas like taxes, personal finance, and compliance. These are industries with well-defined parameters and robust datasets that, if properly integrated into an LLM, can significantly enhance a fintech’s capabilities in acquiring, underwriting, and serving customers.

Parcha is developing AI Agents designed for enterprise use. These agents have the capacity to seamlessly automate manual tasks within compliance and operations by utilizing pre-existing policies, procedures, and tools. Parcha’s ultimate goal is to establish a platform that enables any business to create AI Agents, thereby intelligently streamlining repetitive manual processes. Parcha started its business in the fintech sector because fintech companies’ operations and compliance teams are bogged down with time-consuming, repetitive manual workflows, which puts a ceiling on rapid customer growth. Manual reviews can also be a significant driver of churn as customers are inconvenienced by inconsistent access to financial services, driving down activity rates on the fintech’s platform. Parcha’s AI Agents are digital employees that can take over these repetitive workflows.

JustPaid.io is an AI-powered solution for financial automation invoicing and bill pay. The emergent alignment of end-to-end accounting processes with generative artificial intelligence tools is expanding the capacity of CFOs and finance departments to do more with less. JustPaid collects and summarizes data from contracts all in one dashboard and compares it to both historical and ongoing spending. It can integrate with existing bill payment solution or offers a proprietary multi-approval workflow.

TaxGPT is your AI tax assistant that responds instantly to questions and provides detailed, easy-to-understand answers. The AI model is trained on tax laws and regulations, ensuring accuracy. The main difference between TaxGPT and ChatGPT is the API call to a vector database. TaxGPT has already garnered over 3,400 users and 50,600 questions.

Fintechs’ fight against AI-driven frauds

The advent of easy-to-use AI tools and real-time payments in the US has led to a surge in fraud. This poses significant questions about liability and prevention for consumers and businesses. Generative AI tools can make scams faster and more sophisticated by making it easier, for example, to send out phishing messages, to create a trail of digital activity that seems like a real person while using a manufactured identity, or to duplicate someone else’s activity in order to impersonate them and gather more sensitive information.

There has been a swift embrace of AI and machine learning by financial institutions to combat fraud. According to the benchmark report jointly published by PYMNTS and Hawk:AI, 66% of financial institutions are already employing AI and machine learning technologies, marking a substantial increase from the 34% reported in 2022. What’s more, 48% of financial institutions are either in the midst of integrating or have intentions to adopt AI- and machine learning–powered systems within the next year.

CertifID primarily develops products to fight wire fraud. It is a rules-based engine that utilizes an AI model trained on “internally vetted data,” “expert decisions,” and reviews of its own historical decisions that powers payment disbursement and identity verification processes. The model evaluates various markers of fraud, incorporating new data points as malicious actors embrace new approaches.

Deduce is the only patented technology platform designed to prevent AI-generated identity fraud. Detecting AI-generated identities requires a massive identity graph that sees the majority of the online US population several times per week. Deduce leverages multi-contextual activity-backed intelligence to help organizations identify Super Synthetic identities at account opening and “sleeper” identities already onboarded.

Chargeflow offers a fully automated chargeback management system. Its alerts tool notifies amerchant as soon as a customer files a dispute on a transaction, enabling the merchant to choose whether to proceed with the transaction, offer a refund, or gather more evidence. To maximize the chances of a merchant win, Chargeflow matches incoming disputes to a business’ order data, checks the code and its associated requirements, and generates a dispute response using custom-made templates. Before submission, human experts review the AI-generated responses to ensure accuracy.

Haymaker portfolio with AI-powered services:

AI Insurance (Q1 2022)

AI Insurance offers “policy management, insured management, claims, and business analytics” for modern specialty and captive insurance providers. Currently most specialty and captive insurance companies manage their business out of Microsoft office. AI Insurance’s digital platform provides these companies with a SaaS platform to efficiently process claims, onboard new customers, and underwrite new policies.

AI Insurance is revolutionizing the fundamental infrastructure of an industry plagued by inefficiency and underserved by current solutions. In the United States, there are over 10,000 small insurance companies, with tier 4 and 5 companies alone managing a total of $300 billion in premium. The typical insurance carrier targeted by AI Insurance collaborates with five or more vendors for various functions, such as accounting, claims management, brokerage, and underwriting. Astonishingly, these companies still rely on spreadsheets for handling 90% of their underwriting and claims management processes. Consequently, small carriers find themselves allocating as much as 62% of their income toward administrative costs.

The AI Insurance product is a cloud-based, scalable SaaS solution that allows affordable entry for small and medium-sized insurance carriers. Its single management platform digitizes workflows and promotes collaboration between the vendor ecosystem and the insurance captives. It has been an early visionary for the potential use of AI in the insurance industry, incorporating AI into its current products and building out an aggressive roadmap for AI in almost all of its core systems.

Bookkeep (Q4 2021)

Bookkeep.com is a Brooklyn, New York–based start-up that has developed a smart accounting automation platform for small and mid-sized omni-channel businesses. The platform leverages technology to provide services to automate transactions, track expenses, reconcile bank accounts, file taxes, and create recurring reporting used to better manage financial performance.

Bookkeep.com introduced its SaaS product tailored for small and medium-sized businesses and accounting firms specializing in the retail and hospitality sectors. The product initially made its debut on app stores in September 2019 and secured its inaugural paying customer in February 2020. Subsequently, the company has extended its partnership agreements with merchants and bookkeepers.

Bookkeep’s vision is to become the primary provider of automated bookkeeping by forging integrations with various front-end platforms and accounting software. These integrations are designed to streamline data entry and reconciliation, empowering SMBs to expand by alleviating the burdens of manual data input. Moreover, the product offers real-time financial insights and notifications, catering to evolving market conditions. It features a cost-effective annual pricing plan based on accounting entries, along with a pay-as-you-go option suited for seasonal businesses.

While Bookkeep is integrating AI into its product roadmap, the critical insight is that automating the base level of information so that it is accurate in real time will be critical to allowing AI to be a co-pilot for SMBs in their real time decisions as they allocate expenses and decide on strategic initiatives.

Flyr Labs (Q1 2021)

Based in San Francisco, FLYR is a company dedicated to the development of airline and transportation software geared toward precise demand and pricing forecasting. Its enterprise software platform, known as a “revenue operating system,” harnesses deep learning techniques to enhance airfare forecasting, offering the potential for increased revenue compared to traditional revenue management systems, particularly relevant for industries with high overhead and fixed capacity.

FLYR’s core technology, driven by exclusive machine-learning algorithms, unifies transportation data, providing aviation and transport providers with the means to automate pricing and boost revenues.

In addition to its core offering, FLYR is perfecting an AI-enabled business intelligence toolkit and dashboard designed to enhance decision-making processes for airline revenue management analysts and executives. Beyond airfare price predictions, FLYR can offer contextual analytics, including insights on sellout risk and historical price data. These capabilities enable airlines to better forecast price elasticity and demand for seats, ancillary products, and bundled offers.

FLYR distinguishes itself from traditional airline revenue management systems in two keyways: firstly, through the application of AI-powered deep learning algorithms that assimilate a diverse range of data to enhance forecast accuracy, and secondly, by employing a SaaS delivery model.

FLYR’s approach involves the utilization of an extensive array of airline-related data, encompassing aspects such as bookings, financial data, searches, capacity, and more, in addition to non-traditional data sources like weather and promotional information. Furthermore, it incorporates competitive intelligence, including competitor schedules and capacity data from third-party sources, to inform its modeling process.

While the primary emphasis is on revenue management, FLYR’s AI-powered deep learning models possess the versatility to leverage these data sources for driving ancillary revenues and providing valuable insights to various other departments, such as marketing and planning.