Leveraging AI in Credit Underwriting

Over 100 million Americans are considered financially under-served or subprime, as they've historically been marginalized due to a lack of creditworthiness in the eyes of conventional banks.

For these 100 million Americans, fintechs have been leading the charge in expanding financial inclusion on an unprecedent scale, by introducing artificial intelligence (“AI”) in credit underwriting and customer servicing and by leveraging existing and new data sources in innovative ways. Yet, the promise of full financial inclusion can only be realized by building upon a foundation of trustworthy data, using AI responsibly, and using sound credit underwriting practices.

The promise of AI in driving inclusion. Since our team started using AI in credit underwriting over 13 years ago at Zest.ai, we’ve seen firsthand the potential of financial inclusion at scale by leveraging AI. Specifically, we’ve seen the following to be the most high-impact use cases of AI in subprime financial inclusion:Better risk predictions based on available credit data.

The subprime population tends be overlooked by traditional financial service providers simply because they have a low credit score. AI promises to challenge that paradigm. AI algorithms can identify patterns and trends within data to predict credit risk more accurately. This enables lenders to lend to individuals who may have a higher likelihood of repayment, even if their credit scores may indicate otherwise; further, AI-driven pricing models allow lenders to offer loan products at more affordable rates given the tool’s better ability in predicting repayments.

In short, even staying within traditionally available credit data, AI empowers more credit extended at more affordable rates, thus improving financial inclusion.Better coverage based on alternative dataA big portion of the subprime population are credit invisible, i.e. people who do not have any record or fresh records at major credit bureaus.

Fintech pioneers that have leaned into alternative data have shown initial success in lending to the credit invisible. For example, credit building fintechs are helping the invisible build a credit track record via alternative channels (utilities, rent, etc.), and alternative data such as cash flow data and tradeline data are increasingly being used in making credit adjudications. AI models thrive on more data and more kinds of data; therefore, traditional credit data paired alternative data have and will further improve the power of AI models and drive toward broader financial inclusion.

Unlocking new market segments with faster and smoother processesA big reason traditional lenders stay away from subprime population is the perceived thin profit margin from this cohort. Granted, higher default rate contributes to lower profits; however, high overhead cost also has made many lenders stay away. The profit calculation is changing thanks to the advancement of AI-assisted tools throughout the underwriting and customer service process. Some trailblazers in this area include solutions that allow lenders to easily access hundreds of AI-powered identity verification tools in the digital customer onboarding journey, and solutions that speed up customer servicing via virtual chatbots and intelligent call-routing.

A more cost effective process means that previously unprofitable cohorts may now look like profitable new market segments, thanks to AI.Organized data unlocks faster financial inclusionThe transformative power of organized data in driving inclusion cannot be overstated, but is often overlooked by financial institutions.Data is the lifeblood of the financial industry, especially for first movers that are deploying data-heavy AI models.

Well-structured data catalyzes rapid deployment of AI; on the flip side, we’ve seen many financial institutions struggle with messy internal data. Messy data attacks both the top line and the bottom line. Specifically, data scientists who are supposed to build AI models often end up searching for and cleaning data for weeks, delaying or even killing the rollout of AI models; business leaders who are supposed to run the business are drowning in endless data searches to respond to compliance data requests. What’s worse, in the lending context, inaccurate or missing data often leads to fair lending rule violations, fines, or - in the worst case - business license revocation. In short, messy internal data slows down the industry’s march toward full financial inclusion.Thankfully, several solutions are emerging to automate data organization.

Tools now exist to categorize transaction data and/or cashflow data, making them readily useful for analytics. Further, other tools are emerging to automate a company’s entire data orchestration workflow, from data connection, data mapping, to on-going data pipeline maintenance; these tools free up precious internal resources from mundane tasks and refocus them on what truly matters: building and deploying AI models that improve financial inclusion.

Organized data will propel the whole industry into a new era of accessibility and opportunity.Use explainable models to accelerating financial inclusionPerceived un-explainability of AI models is the biggest hurdle in adopting AI in underwriting, according to a recent Forrester survey. Indeed, financial regulators such as the Consumer Financial Protection Bureau (CFPB) have zoned in on using AI responsibly in issuing credit denials.

We at Ensemblex see a firm path toward AI-fueled inclusion. Below is the framework on how we approach explainability, building upon our experience developing explainable AI models that consistently pass rigorous validation and reviews by government regulators, including ECOA compliance:

Start with approved variablesLenders can always start with a list of variables that are FCRA-compliant and intuitive predictors of credit.

Reduce complexity

Oftentimes, only a handful of variables deliver the bulk of the predictive power. Removing less effective variables will reduce the complexity of the model - and potentially the cost - while making the model easier to explain to internal stakeholders as well as regulators.

Train the team to explain the model

It’s important to find an AI modeling team or external partner that is comfortable explaining the model in plain language. A culture of transparency and trust is key to the long-term adoption of AI.

Balance advancement with understanding

Only use modeling techniques that your team is comfortable explaining. Expand your business’ modeling technique at a pace that’s best fit for your business.Go steady to go farIn lending, there is no silver bullet that will drastically eradicate financial exclusion overnight. The value of new business models and new data takes time to prove out, and it takes even longer for the industry to get onboard. Fintechs in the lending space are playing the long game.With initial success in expanding access to credit, fintechs have an unprecedent opportunity to lead the charge in showcasing financial inclusion.

On the flip side, fintechs that deviate from sound credit management are likely to flame out, leaving the most vulnerable population lacking access to credit, and creating distrust among all stakeholders.

We are optimistic that financial inclusion is within reach, as long as the industry - fintech and traditional players alike - builds businesses on a foundation of organized data, responsible use of data and AI, and sound credit underwriting practices.