Senior Data Scientist, Predictive Modeling, Credit & Risk

About Carvana...

At Carvana, we’re changing the way people buy and sell cars. With an ambitious vision and a fundamentally different approach designed to be fun, fast, and fair, Carvana became the fastest-growing automotive retailer in history. We expanded nationally, went public on the New York Stock Exchange, sold our 1 millionth car, and reached the Fortune 500, all in just eight years.

Today, with 4 million retail customers and counting, Carvana is both the fastest-growing and the most profitable public automotive retailer, and we’re just getting started. We continue to raise the bar for our customers as we tackle the enormous opportunity still ahead in the largest consumer vertical.

Working here means being part of a team that embraces change, celebrates creative problem solving, and always strives to be better. At Carvana, you’ll have the opportunity to take on meaningful challenges, learn quickly, and help shape the future of automotive retail. If you’re driven to grow and make an impact as part of a collaborative team, you’ll fit right in. Learn more about what it’s like to work here from the people that already do. 

Work Model: This is a 100% on-site role at our Tempe office, Monday through Friday.

About the Role

This role sits within Carvana’s Credit & Risk modeling space, working on high-impact predictive models that inform risk assessment and decisioning across the business.

We are looking for a senior individual contributor who will help advance the technical capabilities of our core credit and risk models by developing, validating, and productionizing new modeling techniques and data signals.

Success in this role comes from steady, compounding improvement—introducing new ideas pragmatically, proving value early, and iterating toward more sophisticated approaches over time. While the work will focus on flagship credit and risk models, successful techniques and signals are expected to scale outward to other models and teams through shared workflows and modeling infrastructure.

This role is well suited for someone who operates comfortably at the intersection of advanced modeling and real-world delivery to balance tradeoffs between complexity and business objectives.

What You’ll Be Doing

• Improve core credit and risk models through a sequence of incremental advancements in accuracy, robustness, and coverage over time.
• Leverage a wide range of structured and unstructured data sources across multiple modalities to drive sustained improvements in model accuracy and decision quality.
• Design, train, and deploy advanced machine learning models, including (but not limited to) gradient boosting, representation learning, embeddings, and transformer-based approaches.
• Use high-impact models as a testing ground for new techniques and data modalities, validating which ideas deliver measurable lift in production.
• Build and apply rigorous evaluation frameworks (offline validation, backtests, simulations, online experiments) to guide iteration and decision-making.
• Exercise strong judgment about model complexity and tradeoffs, balancing sophistication with reliability and maintainability.
• Partner closely with data engineering and platform teams to ensure models are production-ready, scalable, monitorable, and maintainable.
• Translate successful work into reusable patterns, abstractions, or signals that can be adopted across other modeling efforts.
• Serve as a technical leader within the Predictive Modeling organization through design reviews, code reviews, and informal mentorship.

What You Should Have

• 5-8+ years of experience building and deploying predictive models in production environments.
• A demonstrated track record of delivering models and improving them iteratively over time, not just developing them offline.
• Strong experience with modern machine learning techniques (e.g., LightGBM/XGBoost, neural networks, representation learning).
• Excellent statistical intuition, including comfort reasoning about bias/variance tradeoffs, generalization, and experimental validity.
• Fluency in Python and SQL, with production-quality coding standards.
• Proven ability to take ambiguous, open-ended modeling problem