In 2007, Steve Jobs launched the iPhone. No one at the time anticipated how transformative the technology would be for banking and financial inclusion. As smartphone ownership grew over the next few decades, the smartphone applications revolution allowed banks to deliver their experiences to populations that historically were hard to reach. Today, mobile banking is the primary means of access for most of the population. Unfortunately, the mobile revolution wasn’t enough to provide financial inclusion for all. A high number of consumers remain out of the financial system.
According to a recent Federal Reserve study, 5% of the US population is “unbanked,” i.e., individuals who lack a checking or savings account with a bank and rely on alternative financial services like cash or prepaid cards. Another 13% of Americans are considered “underbanked,” meaning they have a bank account but also use alternative financial services like payday loans or money orders. In addition, 62 million US citizens have thin credit files, i.e., a limited credit history that prevents them from getting favorable and secure terms with financial institutions for key transactions such as securing a loan or opening a bank account.
With the release in recent years of new large language models and more advanced machine learning tools, there’s a new chance to turn the tide. In the next three to five years, innovative AI solutions will be deployed to solve many structural problems that have held consumers back from financial opportunities.
How AI can help those with thin credit files
Historically, credit bureaus have struggled to provide credit scores for populations with thin credit files. Traditional credit scoring depends on a depth of historical data to give insights about consumers for large, life-changing purchases such as buying a house, renting an apartment, and buying a car. Rich Credit Reports are a key building block of the financial infrastructure powering financial institutions’ flows today – from marketing outreach, anti-fraud solutions, and, perhaps most importantly, Credit Scoring. When confronted with a thin credit file consumer, Credit Reporting Agencies (CRAs) will default to a conservative approach and often return a low credit score, denoting an increased risk for banks and other financial institutions. These consumers end up with unfavorable terms in their loans or mortgages, further preventing them from financial inclusion.
With the rise of AI, there’s an opportunity for Credit Reporting Agencies and industries leveraging credit scores to revisit and improve their credit scoring models. Faster-than-ever training models and hardware already support larger and more diverse datasets. Credit Report Agencies can incorporate insights from many more industries than traditional banking to fix the thin credit file problem. For example, people with thin credit files are more likely to use Buy Now Pay Later (BNPL) products and services and will show similar behaviors to regular lending products that can be used to enhance the Credit File. Similarly, this same group of consumers is more likely to hold gig-economy jobs from platforms like Uber, Lyft, and Doordash for years. The steadiness of employment and income levels across years and billions of transactions will be a treasure trove of information to build more complete Credit Reports.
These industries have large volumes of small transactions spanning years or decades, and the nature of these transactions differs from standard loans. For example, a BNPL loan for a few hundred dollars should not be treated like a standard loan line item for a large purchase and, thus, not impact credit scores and credit files the same way. When CRAs move forward with adding these new data sources, these new data types will strain traditional credit scoring algorithms, but new hardware and software will help tackle these challenges and enable fairer scoring for tens of millions of consumers.
Increasing financial literacy through AI agents
A lack of education about financial literacy is often cited as a key factor in financial exclusion and keeping people underbanked. Forums like Reddit have become famous for modern-day, frequently disastrous adventures where confused consumers seek advice on investing. This kind of outcome shouldn’t surprise anyone working in the financial sector. The modern financial landscape has grown so complex that large portions of consumers fail even the basics of financial literacy. Fewer even use more complex but relevant financial products available to them. Robo-investors in the 2010s made progress towards broader access to financial products, but they are still considered out of reach for the underbanked and unbanked.
The release and rise of ChatGPT and other AI assistants can educate the millions of consumers left behind in the financial sector. AI assistants (also often called AI Agents) are deployed by businesses when humans need assistance, such as customer support. The power of chatbots lies in customizing the answers and experiences for each individual. Modern chatbots and AI technologies now boast “context windows” of hundreds of kilobytes to multiple megabytes. These breakthroughs in large context windows allow modern chatbots to consider a complete picture of customers’ financial history and preferences and deliver tailored help.
Financial institutions can emulate the work done by thousands of startups who have deployed these AI Agents alongside their mobile or web applications. An always-on, always-in-a-pocket AI agent combining banks’ deep expertise and the individual circumstances of consumers will be a powerful accelerator of financial education and, ultimately, inclusion. In the future, every customer will get an AI financial advisor who knows everything about them and can deliver relevant answers to questions such as “Given my current financial situation, how much should I set aside to buy a house for my family in the next five years?” or “What kind of product should I invest in to be able to retire comfortably at 55 in my area?”
That always-on AI agent can play an even more foundational work in financial inclusion by delivering on-time advice to consumers. Studies show that many underbanked individuals stay underbanked because of bad decisions, such as not keeping a high enough balance over time. An always-on AI bank agent will act as a financial coach, giving real-time advice to customers to help them avoid critical mistakes that keep them in the red.
Tailoring financial services to better meet cultural and regional needs
Underbanking statistics show that the rate of underbanking varies significantly by race and ethnicity – as much as a 10x difference between communities. Banks could solve this problem by using head-first, local, and even community-level customization of customer and community messages. Instead, over the last 30 years, banks have had to do the opposite to save on cost and scale. As the industry consolidated, there were fewer and fewer differences in what banks would tell different communities and geographies. This move resulted in strong cost-savings in the industry but also turned many customers away, with people preferring local credit unions and community banks that deliver a more tailored and local message to the community.
Current and future waves of Large Language models (LLMs) reignite the hope for financial institutions to deliver local and community-level messaging cost-efficiently. Thanks to modern large language models, it’s possible for larger banks to customize bank messaging down to specific communities. Financial institutions will be able to deploy Large Models alongside their outreach and marketing efforts to ensure that they can create messaging relevant to particular demographics and age groups. Beyond AI Agents, for Marketing, AI will mark the advent of real-time personalization of marketing messages sent by banks to resonate deeply with every person.
Where do we go from here?
From Superintelligence to replacing knowledge workers, AI brings hope and uncertainty across many industries. There’s an opportunity in the financial industry to use artificial intelligence and large language models to increase financial literacy, reach disenfranchised communities, and give a second chance for people with thin or no credit files. Like the mobile banking revolution that started slow, only to pick up steam and dominate later, not everything from the first batch of AI and LLM solutions will be a home run. The industry needs to continue iterating and evolving its tools and deployments to complete what the mobile revolution in banking started a few decades ago.
