AI advancing inclusive lending?

Abstract: AI has the potential to revolutionize inclusive lending, addressing challenges faced by immigrants and minorities who lack credit history. Traditional credit models often exclude these groups, making it difficult for them to access loans for homes, cars, or businesses. AI can analyze alternative data sources, such as phone records, to assess a borrower’s capacity and character, providing a more inclusive approach to credit scoring. Despite risks like privacy concerns and algorithmic biases, implementing AI-driven models responsibly could break down barriers to credit, fostering financial inclusion and growth for underserved communities. This could significantly benefit both borrowers and forward-thinking lenders.

James is a young immigrant from Africa, recently arriving in Canada with his wife and young daughter. He is ambitious and cannot wait to buy his first home and build a better future for himself and his family. However, something stands between him and his drive and ambition: he lacks the credit history to immediately get a mortgage or buy a good car. Despite having an extensive credit history before arriving in Canada, the slate is wiped clean and he how has no credit history to rely on when he speaks to a lenders.

James represents not one, not two, but millions of immigrants and minorities who may not immediately have access to the one important factor that can help them build a better life: access to credit. We always talk about how credit is important but what makes it so important, and why should credit should be more inclusive?

Credit is chicken-and-egg

Most developed nations in Europe, North America, and more recently, Asia have mature institutional and consumer credit ecosystems. In these countries, based on your credit score you have access to mortgage financing and car loans. Companies have access to business loans and various incentives and credits from government agencies.

United States and Canada, for example, have significantly advanced education and research. There is a track record of students having access to educational loans. With this, they can access great schools and, by extension, a great life. The tuition that comes from funded students becomes the foundation of universities being able to fund research and quality education. With quality education, you see successful alumni who make it in life, who then come back to provide altruistically to endowment funds.

The requirement for credit history is so deeply rooted that it is extremely difficult to break out of this without a great cost to the potential borrower. This is because of the chicken and egg problem. You need a credit history to get good credit. But then credit history can only come from getting credits. It usually excludes those without a good credit history. And those excluded are usually minority groups, the poor, and immigrants.

Credit histories are not to blame

It may be easy for a borrower without an understanding of finance to challenge why there are such strict credit history requirement from lenders and underwriters. Historically, we evaluated borrowers based on the 7 Cs of credit: character, capacity, collateral, contribution, control, condition and common sense. When applying for a loan, the creditors tried to trace your capacity to pay back as well as your character. With the emergence of the credit bureaus, the models and algorithms that help lenders predict who would pay back loans rely on Five C’s of Credit: Character, Conditions, Capital, Capacity, and Collateral and main determinant of character is the credit bureau’s credit score.

… and this is where everything goes south because minorities and new immigrants don’t have enough credit history or anything else to prove their creditworthiness. The traditional scoring model for credit is only as good as the available credit data to use. Bar that, the answer to new borrowers is “no.”

These models are a proxy, predicting the likelihood of loan repayment, it means that it’s not the credit history that is really what’s important, but the assessment of capacity and character. We could assume that if an underwriter has an assurance of capacity and character, in theory, they could extend credit to individuals with thin or non-existent credit history: This is where AI comes in.

AI to the rescue

Over the last 10 years, there have been several studies and experiments about the ability to accurately model and predict a borrower’s capacity and character using alternative data which are not credit history. One of such schemes includes using telephone call data to determine a borrower’s capabilities. Unfortunately, alternative data are unstructured and messy. Sometimes they are not immediately available for every scenario.

The recent rise of AI, including massive advancement by large language models (LLMs) that are built on unstructured data, has shown that significant inferences and intelligence can be gotten from unstructured data which may include video, audio, and text.

An AI-powered credit scoring system could, if it works properly, open massive opportunities for minorities and the poor to have access to credit to build a better life. But even more importantly, it represents a significant growth potential for forward-looking lenders seeking growth and expansion.

Companies like Lendsqr, working with the Federal Government of Nigeria, for example, are building AI models that could use alternative sources of data, including video and audio, to help the poor and vulnerable have access to credit. What we have learned is that:

AI has significant risks

Despite the progress made, many of the AI systems are black boxes which some of the smart engineers and scientists cannot decipher how these results are generated. Many have shown very strong negative biases built in them which can be quite damaging at scale. The issues of privacy cannot be underemphasized because AI may want to see how you speak, check your emails, and look through things that the traditional underwriters will not allow.

AI may also not be a cure-all.

If not implemented properly, we might just end up with a significantly impaired underwriting process, which in turn would lead to catastrophic results. It’s also possible that people could change their behavior just to meet the kind of requirements of the AI system, thereby exploiting its training bias, something that might be difficult to do for other traditional scoring methodologies. We have heard different stories of how AI models are being jailbroken to do terrible things not initially planned by their creators.

However, the risks of not having alternative methods, such as AI-powered models using non-traditional data sources for loan scoring, are real and damaging. It would be beneficial to create these models and start experimenting in smaller scopes before ramping them up to breaking the generational barriers to credit.

We must pilot these AI-driven credit scoring systems in controlled environments, work with regulators to establish clear guidelines and standards, and commit to ethical AI development. Transparency and consumer education are key, as is encouraging collaboration between financial institutions, technology companies, and community organizations.

AI has the potential to revolutionize access to credit and create a more inclusive financial system. The benefits could be enormous, but only if we proceed with caution and responsibility. The stakes are high, but the potential rewards are higher.

James, and many like him, deserve nothing less.