Staff Data Scientist - Credit and Lending ML

About a week before Hurricane Sandy made landfall, the two leading forecast models flatly disagreed: the American one carried the storm safely out into the Atlantic, the European one turned it hard left into New Jersey. The European model wasn't running a better simulation, it was running an ensemble. Hundreds of thousands of people were evacuated on the strength of the spread in forecast distribution, days before any single forecast could have justified moving them, and on October 29th, 2012, Sandy came ashore almost exactly where the models had put it.

At Mercury, we're building the future of financial infrastructure for startups and growing businesses. We're extending credit to those businesses, and we're doing it with proprietary access to some of the best data for understanding a borrower's business, from bank transactions to financial reporting to payroll. We're looking for a Senior/Staff Data Scientists to turn that record into a lending decision, partnering closely with Credit, Product, and Engineering.

In this role, you'll build and own the inference layer behind our small business lending program; you'll ship it into production with the monitoring and documentation that make it trustworthy; and you'll help Credit and Product decide what we should be lending in the first place.

Here Are Some Things You'll Do On The Job:

• Build and own the models behind our lending program - transaction classification, recurring inflow/outflow detection, annual revenue estimation, forecasting with seasonality, anomaly detection, and more
• Turn messy data streams - bank transactions, storefront sales, and more - into a defensible view of a business's health
• Deploy models into a production environment with observability, monitoring, and tests
• Partner with Credit on underwriting policy: what data is asked for, when follow-up information is worth the friction, and how to quantify uncertainty
• Partner with Product on the shape of the product: from principal sizes and pricing to repayment terms
• Assess where predictive modeling actually moves the underwriting program
• Set the technical bar for ML in Credit

You Should:

• Have 7+ years building ML models
• Implemented models that shaped automated decisions with money on the line
• Write production-grade Python
• Have depth beyond classification models; for example: time series and forecasting, anomaly detection, and NLP models
• Be proficient with deep learning, transformers, and LLMs
• Be strong in SQL and at home in a data warehouse
• Be able to explain a model to a credit officer, a model validator, a PM, and an engineer
• Be able to set modeling standards for your data-science colleagues
• Have a quantitative degree, or the equivalent built some other way

Nice To Have:

• Direct experience modeling on bank transaction data
• Experience in small business lending or business credit cards
• Experience with revenue-based financing, merchant cash advances, or similar products

#LI-AC1