Applied Scientist III, Demand Technology, Amazon Demand Side Platform, Bravo Non-endemic

Amazon; New York (NY); On-site; USD 180,000 - 240,000; Full time

Job summary

Amazon's Demand Side Platform (DSP) team is seeking an Applied Scientist III to own the science behind conversion prediction, bidding, ranking, and measurement for non-endemic advertisers.

You will design ML models, define measurement frameworks, and work with cross-functional teams to deploy models in production, shaping outcomes for advertisers worldwide.

Responsibilities

• Own the applied-science roadmap for non-endemic performance workstreams—from problem framing to offline/online validation and production launch.
• Design and improve ML models for sourcing, ranking, and response prediction to optimize CPA and ROAS.
• Define measurement methodology (incrementality, weighted conversions, off-Amazon attribution) to value impressions with off-Amazon conversions.
• Collaborate with engineering, product, and sales teams on deep dives into enterprise accounts to translate findings into scalable improvements.
• Mentor scientists and engineers; publish internal best practices and contribute to the broader scientific community.

Job description

Description

Amazon's Demand Side Platform (DSP) helps advertisers reach audiences across the web, and we are tackling one of the hardest open problems in this space: making performance advertising work for non-endemic advertisers — brands in financial services, telco, auto, travel, and direct-to-consumer that sell outside of Amazon. As an Applied Scientist III on the Demand Technology team, you will own the science behind conversion prediction, bidding, ranking, and measurement for these advertisers, where standard signals are sparse and new modeling approaches must be invented. This is a company-level priority with significant room for scientific impact, and the models you build will directly determine whether advertisers see the outcomes that keep them investing on our platform.

Key job responsibilities

• Own the applied-science roadmap for one or more non-endemic performance workstreams — from problem framing and experiment design through offline validation, online experimentation, and production launch.
• Design and improve machine learning models for sourcing, ranking, and response prediction that optimize toward advertiser outcomes such as cost per acquisition and return on ad spend.
• Define measurement methodology — including incrementality, weighted conversions, and off-Amazon attribution — that correctly values an impression when the conversion happens outside Amazon.
• Partner with engineering, product, and sales-facing teams on deep dives into real enterprise advertiser accounts, turning account-level learnings into scalable model and system improvements.
• Mentor scientists and engineers across the organization, publish internal best practices, and contribute to the broader scientific community through peer reviews and publications.

A day in the life

You might start your morning analyzing offline experiment results for a new conversion-prediction model, then join a design review with engineers on how to serve that model in the real-time bidding pipeline. After lunch launch, you could be working with a product manager to define success metrics for a non-endemic advertiser segment, followed by a code review for a teammate's feature-engineering change. You regularly carve out time to read recent research on causal inference or data-efficient learning and assess whether new techniques could improve your team's models.

About the team

We are a lean team of applied scientists and software development engineers within Amazon DSP, focused on strategic initiatives that have not had significant investment before. We work close to the customer — partnering directly with product and sales-facing teams to understand advertiser needs and translate them into scalable scientific solutions. If you want to shape the direction of a high-priority problem space where your ideas move quickly from whiteboard to production, we would love to hear from you.