Principal Data Scientist

Flagship Pioneering, Inc.•Cambridge, MA

3d•$179,000 - $236,500•Onsite

About The Position

As a Principal Data Scientist, you’ll be a technical leader responsible for the full lifecycle of the models and data science systems that predict and optimize advertising performance across channels. You’ll define the learning problems, training data, targets, and business outcomes to influence; train and rigorously validate the right models offline; and take the technology to market through customer-facing validation. You’ll work across paid social, search, display, video, and emerging channels, translating complex, noisy advertising data into robust models, explainable decisions, and measurable customer value as our platform and data scale.

Requirements

8+ years of data science, machine learning, statistics, or related quantitative experience.
Advanced degree in Statistics, Mathematics, Economics, or another quantitative field, or equivalent practical experience.
Strong expertise in Python, SQL, and modern data science and machine learning frameworks.
Strong foundation in statistics, machine learning, experimental design, forecasting, and model evaluation.
Demonstrated track record of taking models from research and experimentation through production deployment, customer validation, and measurable business impact.
Experience modeling advertising, marketing, consumer, or marketplace performance, including metrics such as conversion, engagement, acquisition, ROAS, CAC, or lifetime value.
Experience working with advertising data across multiple channels and platforms, including paid social, search, display, or video.
Experience with causal inference, incrementality testing, attribution, media mix modeling, or related approaches to measuring marketing effectiveness.
Exceptional scientific communication skills, with the ability to earn trust across technical and non-technical stakeholders by clearly explaining assumptions, evidence, trade-offs, uncertainty, and validation results.
Comfortable working in a fast-paced, evolving environment with minimal oversight.

Nice To Haves

Experience building predictive advertising or marketing optimization products in an early-stage startup.
Deep familiarity with advertising platform data and APIs, including Meta, Google, TikTok, LinkedIn, or similar platforms.
Experience with budget optimization, bidding, recommendation systems, uplift modeling, or other decision-making systems.
Experience applying LLMs, embeddings, or other foundation models to advertising, creative, or consumer data.
Experience developing models in data-sparse or cold-start environments where historical performance data is limited.
Comfort with ambiguity and making trade-offs between modeling sophistication, speed, interpretability, and business impact.

Responsibilities

Define Learning Targets & Model Advertising Performance – Identify the customer decision and business intervention to influence; construct training labels and measurement windows; assess data quality, bias, and leakage; and develop predictive models that forecast campaign and creative performance across channels, audiences, placements, and objectives.
Build Cross-Channel Measurement & Optimization Systems – Develop methods to compare and optimize advertising performance across paid social, search, display, video, and other channels while accounting for differences in measurement, attribution, and data availability.
Develop Experimentation & Causal Measurement Approaches – Design and analyze experiments, incrementality tests, and causal inference approaches to verify that the targets we optimize change business outcomes, not just model metrics, and to distinguish correlation from true impact.
Translate Models into Marketable Decisions – Turn model outputs into clear, customer-facing recommendations for campaign strategy, budget allocation, targeting, creative selection, and optimization—decisions that can be explained, tested, and proven useful in market.
Advance Modeling & Validation Best Practices – Shape our approach to forecasting, experimentation, feature engineering, model selection, and production data science, including holdout and backtesting design, calibration, uncertainty, failure modes, data drift, and decision thresholds before production or customer exposure.
Own the Model Lifecycle Cross-Functionally – Partner across data, product, engineering, and customer discovery to move models from learning problem through deployment and customer validation, while supporting the quantitative work needed to make that lifecycle successful.