Marketing Science Internship Senior Project (PFE)

TOPIC 1: Advanced Multicollinearity Mitigation Techniques for Marketing Mix Models

This project focuses on one of the most common challenges in Marketing Mix Modeling: Multicollinearity. Because companies often activate multiple media channels simultaneously media variables tend to share very similar trends. This correlation makes models unstable inflates coefficients and reduces interpretability.

This project aims to:

• Explore the underlying causes of multicollinearity in MMM

• Evaluate alternative techniques used across the industry

• Improve and automate MASS Analyticss methodology to make composite-variable creation more scalable transparent and aligned with marketing logic.
  

Main Competencies:

• Experience with common data science toolkits (Python R etc.)

• Good understanding of econometrics concepts and modeling methodologies

• Familiarity with data processing techniques and best practices

• Strong Python skills

• Analytical and problem-solving mindset
  

Learning Outcomes:

• Develop a structured understanding of multicollinearity and its impact on MMM

• Identify assess and automate key steps in the composite-variable creation process

• Improve the efficiency and reliability of the modeling process through better variable engineering

• Build a scalable application or framework that automates the creation of Transformed Variables following clear business logic

TOPIC 2: Building an Automated Pipeline for MMM Scenario Simulation

Many advertisers want to evaluate how different media activation strategiessuch as Always-On and On/Off patternscould impact future performance. However detailed media plans at the required granularity are often unavailable making scenario construction slow and partially manual.

This project focuses on transforming high-level strategic planning into detailed week-level laydowns that can be fed into MASS Analytics forecasting tools. The objective is to automate scenario generation enable fast comparison of alternative activation plans and deliver clearer insights for decision-making.

You will analyse historical activation patterns design logical rules for converting strategy into weekly laydown and propose a scalable automation component that integrates seamlessly with our forecasting workflows.

Main Competencies:

• Experience with common data science toolkits (Python R etc.)

• Understanding of econometrics concepts and MMM methodologies

• Familiarity with data processing techniques and best practices

• Proficiency in Python

• Strong analytical and problem-solving skills

Learning Outcomes:

• Develop a structured methodology for translating marketing strategies into granular MMM-ready scenarios

• Automate scenario creation to enhance forecasting speed and reliability

• Improve the consistency and scalability of laydown planning

• Strengthening understanding of media planning logic and MMM forecasting workflows

TOPIC 3: Leveraging & Benchmarking Open-Source MMM Approaches to Strengthen Marketing Mix Projects

This project explores how open-source Marketing Mix Modeling platforms and complementary competitive intelligence tooling can be used to benchmark results stress-test assumptions and support MASS Analytics modeling algorithms and storytelling.
The goal is to compare outputs across different MMM philosophies understand why results differ and identify where MASS Analytics methodology is stronger while also spotting areas where open-source tools can add value (diagnostics uncertainty experimentation etc.).

Main Competencies:

• Experience with common data science toolkits (Python R etc.)

• Solid understanding of econometrics concepts

• Strong Excel & Python skills

• Analytical and problem-solving mindset with attention to methodological rigor and comparability

Learning Outcomes:

• Map the open-source MMM landscape and categorize tools by methodology strengths and limitations (e.g. frequentist vs Bayesian automation level support for hierarchical models uncertainty calibration).

• Design a consistent benchmarking framework to run the same dataset / same business question across approaches using aligned transformations and evaluation metrics.

• Compare model behavior and outputs (contributions ROI curves diminishing returns response curves stability under collinearity sensitivity to priors/hyperparameters).

• Build an assessment for Mass Analytics: where our tools are stronger vs where open-source can add value.