Fractional Data Scientist, Marketing Mix Modeling

Role Overview

About the Project

We are developing a privacy-first healthcare marketing attribution platform designed to solve the "tracking gap" created by evolving HIPAA and OCR regulations. As traditional 1:1 pixel tracking and Multi-Touch Attribution become non-compliant, our platform provides healthcare organizations with a server-side, aggregate-data approach to measuring channel performance.

We are seeking a Marketing Attribution Data Scientist to architect the modeling layer of a scalable Media Mix Modeling product. You will be responsible for building the statistical "brain" that translates aggregate marketing and performance data into actionable strategic insights.

The engineering team will handle the serverless data pipelines, AWS infrastructure, dashboard, and UI. This role will focus on the mathematical engine: model selection, prior calibration, model logic, and output structure.

The Role

You will lead the design and implementation of an automated Bayesian MMM framework. Your goal is to create the modeling foundation that helps marketers understand channel performance, identify diminishing returns, and make smarter budget allocation decisions after traditional click-level identifiers are removed.

You will collaborate with the founder, existing full-stack developer, and dashboard designer to ensure the model outputs can be handed off cleanly for dashboard and reporting implementation.

Key Responsibilities

1. Bayesian Model Architecture

• Framework Implementation: Evaluate and recommend a robust Bayesian MMM framework, with preference for Google Meridian, PyMC, NumPyro, or similar. Familiarity with other MMM tools such as Meta Robyn is a plus, but not required.
• Prior Calibration: Develop a methodology for defining Bayesian priors using historical data and, where applicable, localized "micro-holdout" / geo-testing variance.
• Advanced Modeling: Incorporate adstock effects, decay, and non-linear saturation curves to help identify optimal spending thresholds across marketing channels.
• Model Inputs / Outputs: Define the required input schema, expected output structure, assumptions, and limitations of the model.

2. Attribution & Budget Optimization Logic

• Channel Performance Modeling: Build logic to help estimate the contribution of different marketing channels using aggregate, privacy-safe data.
• Privacy-Safe Attribution: Account for server-side tracking, Conversion API context, and aggregate attribution models where traditional click-level identifiers are unavailable.
• Budget Optimization: Define how the model should identify saturation points, flag diminishing returns, and recommend budget reallocations.
• Probabilistic Reporting: Generate model outputs that can support both executive-level median estimates and deeper analysis using confidence intervals.
• Handoff Structure: Package the modeling logic and outputs in a way that can be consumed by the existing developer and dashboard designer.

3. Documentation & Collaboration

• Model Documentation: Provide clear documentation of model assumptions, methodology, priors, limitations, and recommended maintenance approach.
• Technical Handoff: Collaborate with the existing developer on how the model should connect to the broader product environment.
• Insight Translation: Help define how raw statistical outputs should translate into plain-English marketing insights, which can later be surfaced through the dashboard or reporting layer.

Technical Environment

The broader product is being built within a HIPAA-conscious, serverless AWS ecosystem. Direct ownership of AWS infrastructure, Bedrock implementation, and dashboard development is not required for this role. The existing developer will support infrastructure and deployment needs.

Current / planned environment includes:

• Data Lake: Amazon S3 & Amazon Athena
• Compute: Amazon SageMaker or similar model execution environment
• Storage / Cache: Amazon DynamoDB
• Dashboard / UI: Handled separately by the product / dashboard team
• AI Layer: Future-state reporting layer; not part of the core scope for this role

Ideal Candidate Background

• Deep experience in Marketing Mix Modeling, marketing attribution, marketing science, or advanced media measurement.
• Strong experience with Bayesian inference, probabilistic modeling, or applied statistical modeling.
• High proficiency in Python and experience with MMM / probabilistic modeling frameworks such as Meridian, Robyn, PyMC, NumPyro, or similar.
• Experience working with aggregate attribution models, server-side tracking data, Conversion API context, or privacy-safe marketing measurement.
• Ability to translate complex statistical outputs into clear, actionable business recommendations.
• Ability to define model inputs, outputs, assumptions, priors, and limitations.
• Comfortable scoping ambiguous technical requirements in a startup-style environment.
• Strong written communication skills for asynchronous collaboration.

Nice-to-Haves

• Experience with North American healthcare marketing, especially US-based healthcare marketing, is a strong differentiator. Familiarity with healthcare marketing channels, patient acquisition, or regulated healthcare advertising is highly valuable.
• Familiarity with AWS-based model deployment environments, especially SageMaker.
• Experience building models that feed SaaS dashboard or reporting products.
• Experience with geo-testing, incrementality testing, or micro-holdout methodology.
• Experience translating statistical outputs into plain-English insights or executive-facing recommendations.

Engagement Structure

• Phase 1: Scoping, framework recommendation, input/output schema definition, implementation plan, and estimated effort for Phase 2.
• Phase 2: Model build, validation, documentation, and handoff to the development / dashboard team.
• Data Access: Work will be performed in a staging/dev environment with synthetic or de-identified data; no live production PHI access is expected.

Success Metric

A successful engagement results in a defensible Bayesian MMM / attribution modeling framework that provides marketers with a reliable "single source of truth" for channel performance, even when traditional click-level identifiers are removed for compliance reasons.

The model should be clearly documented, practical to maintain, and structured so the existing development team can integrate it into the broader product experience.