Affirm Holdings: supervised machine learning and market disruption

The case study is going to be about Affirm Holdings. Affirm Holdings, now a prominent financial technology company. They specialize in buy now and pay later financing solutions. This company was established in 2012 in San Francisco. They had rapid expansion with significant venture funding and some financial milestone is that it surpassed one billion dollar in funding and it went public in 2021. And it now collaborates with major retailers including Amazon, Walmart, and Target.

However, this company was not always successful. It had to start somewhere. It had to first disrupt the traditional credit model. This is a big challenge because it means entering against the big financial institutions and the big banks. To do that, it had to have a very specific focus. Its focus was going to be addressing the needs of customers that were not served well by these legacy credit systems. For example, this could be individuals with thin credit files or low FICO scores or millennials and others seeking flexible payment options. That is people that the bank would normally not consider credit worthy but that they thought they could actually repay their loans if provided with one. So there was a huge market opportunity. The market opportunity was capitalizing on the gap left by conventional lenders and targeting a segment largely ignored by the traditional banking system. However, they had to overcome a big challenge. Because if you're trying to enter a new credit market segment, you may not have the data to do that. The big business in credit is, I'll give you a loan, I'll charge you interest, but you'll repay me. If you don't repay me, then I'm going to lose money. But if I offer you a higher interest rate because I think you will not repay me but you will actually do, then you may just go to a different lender that gives you a lower interest rate. You have to balance both the interest rate that you're charging against the probability of default. That's an extremely tricky business and normally, banks would rely on established credit agencies to do that. But they wanted to provide new clients without access to these traditional credit markets to lending opportunities. How did they do that? What they did is they turned to machine learning. In particular, supervised machine learning. The nice thing about supervised machine learning is that it allows you to use any characteristic that you have available. For example, maybe you don't want to look at only the credit history, you want to look at the current income and maybe you want to look at the education, and maybe you want to look at whether they live in a community where people normally pay their bills on time. You would use all of the insights into this and try to obtain a better forecast of the probability of default. With a better forecast of the probability of default, then you are able to decide, can I actually lend to this person even though they don't have a credit score? If the answer is yes, you can also tailor the interest rate so that they're likely to get the loan and not default in the meantime.