Build an adaptive pricing system for loans on Upstart's lending marketplace

You are a product manager at Upstart, the online lending marketplace that uses machine-learning models to decide which consumers to approve for loans, and on what terms, for the banks and credit unions that partner with it. You are asked to design dynamic pricing for Upstart. Decide what should be priced dynamically and what the price should respond to, describe how the pricing would work for borrowers and for the partners and investors who fund the loans, and explain how you would judge whether it is working. ```hint Price for three audiences A loan's price has to win the borrower, earn enough for whoever funds the loan, and hold up under the rules on how credit may be priced. Settle what the price is allowed to react to, and what it must never react to, before designing the mechanics. ``` ### Clarifying Questions - Which price is in scope: the rate and fees offered to borrowers, the fees Upstart charges its lending partners, or both? - Which loan product should we start with: unsecured personal loans, auto refinancing or home equity lines of credit? - What matters most: growing the number of funded loans, earning more per loan, or protecting the returns that keep funders on the platform? - Who has the final say on a loan's price: Upstart, or the partner whose credit policy and return targets apply to that loan? ### What a Strong Answer Covers - A clear scope for what is priced and a goal that balances funded volume, returns for funders and good borrower outcomes. - The inputs that move the price (each applicant's risk, the cost and availability of funding, market conditions) and how they combine into an offer within legal limits and partner policies. - Fair-lending, disclosure and explainability requirements, including the inputs the price must not use. - Success metrics across conversion, credit performance and fairness, with guardrails. - Risks such as losses rising above predictions, funding drying up and adverse selection, with a governed rollout and testing plan. ### Follow-up Questions - Your data shows that some approved borrowers accept almost any rate they are offered. Would you use that in pricing, and why? - Recent loans are defaulting faster than the model predicted. How should prices respond, and what are the trade-offs? - How would you test a pricing change safely when giving similar borrowers different rates raises fairness concerns?