Pricing Optimization Scientist, Pricing Science

Take hard sequential decision problems, like delayed rewards, shifting demand and strict business constraints, and turn them into pricing policies that run at Amazon scale. Few places let you test RL ideas on a problem this large.

Key job responsibilities

Every day, Amazon makes hundreds of millions of pricing decisions. Each one is a choice made with incomplete information: demand shifts, competitors move, inventory runs low, and every price we set changes the data we learn from next. The P2 Optimization Science (P2OS) team builds the science behind these decisions. Our work includes demand response models, the optimization systems that turn them into prices, and the experiments that prove those prices help customers and the business.

We"re hiring an Applied Scientist to build sequential decision-making methods for pricing. Most pricing systems estimate demand first and optimize second. We want systems that learn while they decide. That means deciding when to explore and when to exploit, how to trade short-term revenue against long-term customer trust, and how to set safe prices in categories where we have little history. You"ll work on contextual bandits, reinforcement learning and online optimization problems that most research setups ignore: delayed and noisy rewards, hard business constraints, non-stationary demand, and a catalog too big for a separate model per product.

This is an applied role. You"ll take methods from offline simulation through online experiments and into production. Your success will be measured by what changes for customers and the business, not by benchmark scores. You"ll partner closely with senior scientists, economists, software engineers and product managers. You"ll also help shape the team"s long-term direction on how Amazon prices adaptively.

If you like hard optimization problems with real stakes, and you"d rather ship a method that works than publish one that doesn"t, we"d like to talk to you.

A day in the life

As a Pricing Applied Scientist, you"ll shape the prices customers see on a very large share of Amazon"s catalog. In a typical week, you might:

• Prototype a contextual bandit policy for a product segment with sparse sales history, then stress-test it in simulation against the current production approach.
• Dig into why a policy is exploring more than expected and trace it to a reward signal that arrives days late.
• Work through a reward design with an economist: how to credit a price change that lifts conversion today but may affect repeat purchases next month.
• Review an online experiment readout with your product manager and decide whether the policy is ready to scale or needs another iteration.
• Write a short science proposal on a new constraint, like limiting how often a price can change, and estimate its cost to the objective.
• Pair with an engineer to make a policy fast enough to update prices in near real time.
• Present your findings at the team"s science review, or develop them into a paper for a venue like KDD, NeurIPS or EC.

About the team

The Pricing and Promotions Optimization Science (P2OS) team is the Optimization 'Application Layer' within Pricing Sciences. Our team is a small unit of specialized science resources equipped to solve some of the organization's most ambiguous challenges. Our charter is to develop models to set optimal prices and promotions, shared common tooling across all merchants, rule and policy setting to drive error detection and price sharpness, and innovative price guardrails and incentives to support long-term marketplace health.