Consumer switching in T-Mobile/Sprint merger
Analyzed U.S. telecom behavior post-merger to quantify switching rates and drivers.
- Causal inference, quasi-experimental design, machine learning
- Python, R, large-scale data processing, visualization
Hi, my name is Javier Torralba, PhD Candidate in Quantitative Marketing at Tilburg University.
My background lies in Economics, Data Science, and Marketing Analytics, with research experience using methods like causal inference and machine learning. Below you can find my research, projects, values, and how to reach me.
2023-2024
MSc in Data Science and Society
I took courses and specialized in machine learning, deep learning, and working with real-life complex data. I completed a thesis where I predicted Instagram posts popularity using multimodal ML methods. I worked with advanced tools such as PyTorch, TensorFlow, and gradient boost algorithms while developing strong analytical and modeling skills.
2020-2023
BSc in Economics
The BSc in Economics helped me build a strong foundation in microeconomics, macroeconomics, and econometrics. During the program, I developed strong quantitative reasoning skills and worked as a research assistant with professors in the marketing department, contributing to behavioral economics research.
2019-2020
Propedeutic Year in Finance and Accounting
This program provided a strong foundation in financial management, financial accounting, and management accounting. It shaped my economic intuition and continues to inform my thinking and research today, while also sparking a lasting interest in optimizing personal and household financial planning.
About
Data driven researcher with an impact
I blend econometrics, machine learning, and a deep contextual understanding of the problem at hand to explain why people behave the way they do, and what to do next.
What I do
Design research and analytics that quantify complex data, using causal inference and predictive modeling.
How I work
Methodical and collaborative: rigorous methodology, reproducible pipelines, and clear visual narratives.
What I value
Proactivity, communication, and practical impact—making complex analyses understandable for everyone.
A look behind the data
How I think and work with others
Outside of my PhD, I enjoy building small data-driven tools, exploring financial planning, and thinking about how people make decisions in everyday life. I’m especially interested in projects that sit between theory and practice, and in work that helps people make better choices. I value collaborative environments where curiosity, clarity, and creative problem-solving matter.
Applied Projects
Selected work & research
Recent projects spanning research, analytics, and applied data products.
Growth AI Marketer
Built an end-to-end outreach system for Cavi Production to increase agency sales by 60% through personalized, automated campaigns.
- Python, OpenAI API, Selenium for lead scraping at 20k+ scale
- Automated 100k+ personalized emails with tracking
Dutch Apartment Finder
Created a notification pipeline that alerts users as soon as relevant housing listings go live.
- Python, APIs/web scraping, alerting
- Delivered 100+ housing alerts; helped secure 2 apartments
Academic work
Research & theses
Formal research projects and theses that explore consumer behavior and quantitative methods.
Data Science Master Thesis
Using a dataset of 133,642 Instagram posts from 27,893 users, I built a multimodal prediction system combining ResNet50 image features, BERT-based text embeddings, and user metadata to forecast post popularity before publication. The best model (LightGBM) achieved a strong ≈69% accuracy—far above the 50% baseline—and revealed through SHAP explainable-AI analysis that follower count, post history, and subtle image-text interactions drive engagement. An error-curve analysis uncovered that predictions near 0.5 have ~45% error while confident predictions near 0 or 1 drop to ~5%, enabling creators and marketers to make smarter posting decisions using both model output and its confidence
Open PDFEconomics Bachelor Thesis
During COVID-19 lockdowns, using a dataset of 6.5 million Goodreads reading records across 31 countries, I show that stricter restrictions caused a sharp surge in reading behavior. When societies partially shut down, users read 6.7% faster, consumed 6.9% more pages, and finished 6.5% more books, and under full lockdown these effects doubled with users reading 14.5% faster, 11.6% more pages, and 10.4% more books. Readers also shifted toward older titles (+5.7%), while their book ratings remained unchanged, revealing a clear behavioral shift in consumption patterns without changes in sentiment.
Open PDFCapabilities
Tools & methods
Combining rigorous research methods with practical analytics and communication.
Data & Methods
Causal inference, econometrics, Bayesian modeling, forecasting, A/B testing, machine learning.
Programming
Python, R (tidyverse), SQL, APIs, Git/GitHub, data pipelines, AI/LLM applications.
Visualization
Data storytelling with dashboards and visuals; 250+ visualizations for research and stakeholders.
Teaching & Collaboration
Teaching data preparation/programming to 60+ students; clear documentation and cross-functional communication.
Contact
Let’s collaborate
Let’s discuss research, analytics, or collaboration opportunities.