Staff AI Scientist (Fintech Risk AI/ML) at Intuit in Mountain View, CA

Job description

The team Money Movement Risk AI/ML is a team of AI scientists and ML engineers in Intuit's Fintech Risk organization. Our models decide, in real time, whether money moves for small businesses on QuickBooks: card and ACH payments, money coming in, money going out, when funds are released, and how long a payment is held. Every decision has two failure modes. Approve the wrong transaction and Intuit takes a fraud or credit loss. Decline or hold the right one and a small business waits for money it has already earned. Our job is to push both down at the same time. Our mandate: AI touches every money movement risk decision. That covers deciding (authorize, fund, release), acting on the decision (queue, investigate, hold), and accounting for it (expected loss). The role You will be the technical owner of one money movement risk capability: its models, its evaluation, its production health, and its business outcome. You will design, train, validate, and ship the models yourself, and you will set direction for the scientists and engineers who build with you. This role is for an AI producer. You will build new models from first principles and put them in front of live money. It is not a role for wiring an existing LLM into a product surface. What you will work on Depending on fit and team needs, you will own one or more of these areas: Authorization risk. Real-time fraud and loss models that approve, decline, or step up card and ACH transactions, for both money coming in and money going out. Funding and settlement risk. Models that decide when funds are released and how long payments are held or delayed, so good customers get paid faster and risky payments are caught before money leaves. Balance and account-level money risk. Risk on stored balances and on the accounts behind them, as more money stays inside Intuit. Day-1 risk for new money products. Risk decisioning that is ready when a new payment or funding product launches, often with almost no labeled history. Agentic money movement. Transaction risk for payments and money actions initiated by AI agents and assistants: agent identity, consent, and intent. Risk-ranked queueing. Ranking cases for risk operations by dollars at risk, so analysts spend time where the loss is. Responsibilities What you will do Own outcomes, not just models. Define success in loss, approval rate, hold rate, and time-to-funds. Ship models that move those numbers, and own them in production. Solve the hard statistics of risk. Delayed and noisy labels (chargebacks and returns arrive weeks later), extreme class imbalance, and selection bias from models whose decisions shape their own training data. Use holdouts, reject inference, and counterfactual evaluation to measure what actually happened. Build for adaptive adversaries. Detect fraud rings, account takeover, and new attack patterns with graph, sequence, and anomaly-detection methods. Test models against attackers who probe, retry, and adapt. Make models production-grade. Partner with ML engineers on real-time features, low-latency serving, automated retraining, and monitoring for drift, data quality, and loss. Turn Intuit's data into an advantage. Build reusable, versioned features and labels from transaction, account, and network data that other models on the team can build on. Set the evaluation bar. Establish offline and online evaluation, experimentation, and adversarial testing standards for your area. Lead across functions. Work with risk policy, product, engineering, risk operations, and finance. Explain trade-offs between loss and friction in plain terms and drive decisions. Raise the team. Mentor senior and early-career scientists and engineers, lead design and model reviews, and help hire. Qualifications What you bring Required 8+ years building and deploying ML models in production, including significant work in fraud, payments, credit, or another adversarial or financial-risk domain. MS or PhD in Statistics, Computer Science, Applied Mathemati