Data Scientist, Supply

About Anthropic Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems. About Anthropic Anthropic is an AI safety and research company. We build reliable, interpretable, and steerable AI systems, and we believe AI will have a vast impact on the world — our goal is to ensure that impact is positive. About the role Anthropic is compute-constrained, and how we allocate that compute is one of the highest-leverage decisions we make as a company. Today, allocation choices are only loosely tied to the user outcomes we ultimately care about — retention, lifetime value, and the experience of people relying on Claude. This role exists to change that by addressing two intertwined problems at the heart of how we allocate compute. The first is an allocation problem: matching a volatile, heterogeneous stream of demand to a finite, heterogeneous fleet of chips. Which models run on which hardware, in which regions, under what serving configurations — with demand shifting and capacity bounded — is a problem the team navigates continuously today, with more intuition than rigor. You will bring structure to it: building the metrics and analytical frameworks that make the trade-offs legible, and partnering with the infrastructure teams that own these systems to turn that understanding into better decisions. The second is a causal-inference problem: there are many levers — rate limits, pricing, cache behavior, capacity shifts, routing changes — and only a partial picture of what pulling each one actually does to the users on the other end. You will build the causal understanding that closes that gap, choosing whatever approach the question calls for, so allocation decisions are made on expected user impact rather than intuition. This role is a fit for someone who thinks natively in terms of constrained allocation and queueing, who treats "what would happen if we changed X" as an identification problem rather than a dashboard query, and who wants their work to translate into operational and productionized change. You will work closely with the infrastructure engineers who run our compute, and your findings will be presented to senior leadership. Key responsibilities - Build and run testing frameworks — observational and synthetic — to quantify how different inputs affect compute allocation outcomes - Connect compute allocation decisions to downstream user outcomes (retention, lifetime value, revenue) - Partner closely with infrastructure engineers, product, and research to instrument systems, measure what matters, and ship operational changes - Develop the metric hierarchies, dashboards, and reporting that turn supply decisions into shared understanding across the company - Contribute analyses and recommendations to executive forums, and co-author the supply narrative shared with the CTO and staff Minimum qualifications - Strong technical individual-contributor background in data science, analytics, or operations research - Demonstrated comfort reasoning about resource allocation and trade-offs under constraints — drawn to systems problems, not just dashboards - Working fluency with causal inference — able to recognize when an effect needs to be identified, not just measured, and to choose an appropriate design - Deep proficiency with Python, SQL, and data visualization tools - Track record of owning analyses end-to-end and communicating results clearly to engineering and product leadership - Direct experience working closely with engineering teams on production systems - Alignment with Anthropic's mission of building helpful, honest, and harmless AI Preferred qualifications - 8+ years of hands-on data science experience - Significant technical individual-contributor experience in data science, analytics, or operations research at staff level scope - Experience with highly complex systems with many interacting components (ad networks, payment processing, marketplace matching, routing, etc.) - Hands-on operations-research depth: experience formulating and shipping real-time constrained-allocation, routing, or scheduling problems in production (LP/MILP, queueing, or RL-based control), with the ability to defend modeling choices - Causal-inference depth beyond off-the-shelf quasi-experimental templates — particularly methods for recovering long-term impact from short-horizon data: surrogate/proxy-outcome models, off-policy evaluation and counterfactual policy learning, or structural approaches, built rather than merely run - Experience contributing to or designing experimentation platforms, not just using them - Exposure to AI/ML products, large language models, or large-scale inference systems - Track record of setting technical direc