Can you explain what you mean by circumventing state rate caps?

Each state has its own set of lending APR caps. If you go through a chartered bank, typically in Utah, they have a term called "rent a charter" where you partner with a bank to use their national charter for loans. This allows you to bypass individual state laws on lending. For example, if New York has an APR cap of 9.99%, you don't have to follow that cap. You can go above 36%.

How do you train this model, whether built in-house or purchased? Do you use alternative data on subprime borrowers, or only past data from loans you've issued?

Historically, it's been both, to be honest. But I think the issue really is that I've spent a lot of time working with our model teams over the years. What I've learned is that the data can become stale more quickly than you might think. So, I believe the model needs to be constantly updated, especially now with machine learning models, which are continuously retrained. Even where I work today, it's a constant iteration. Some companies are more nimble than others in learning how quickly to tweak the model based on external factors, like market rates versus what we're offering and the cost of borrowing money to lend. All these factors contribute to profitability. The margins are thinner as you go higher in the credit spectrum because a 700 FICO borrower expects a better rate. It's a constant evolution. We used to purchase data, but there was a time when our data was five to seven years old, and it wasn't as useful anymore.