41k views
Upgraded Marketing Mix Modeling
A deep dive into modern MMM techniques, moving beyond simple regression into Bayesian approaches.
Senior Staff Data Scientist · PhD Mathematician · Educator
Dr. Robert
Kübler
I find the algorithm hiding inside a vague operational problem, then take it from proof to production. Mostly forecasting, causal inference, and Bayesian methods. PhD in mathematics. 8+ years across research, engineering, and teaching.
- Forecasting savings
- €15M+
- Production time series
- 500+
- Article reads
- 1M+
- Professionals taught
- 50+
About
Making the people around me better
I care deeply about sharing what I know.
I designed and deliver the official sktime forecasting curriculum for the German Center for Open Source AI, now in its third cohort, and mentor data scientists directly wherever I work: five university theses supervised during my PhD, plus twelve years volunteering as a math tutor. Alongside that: 90+ articles on Medium for a 1M+ readership, talks at data science meetups (Zalando, inovex), and shipped features to open-source libraries including scikit-lego, sktime, and PyMC-Marketing.
Experience
Roles & impact
Employment
Senior Staff Data Scientist
- Set the technical direction for a supply chain forecasting product optimizing warehouse staff planning, and drove sktime adoption into 2 further teams beyond my own
- Replaced the prior planning process with sktime, cutting offline MAPE by 10%. Now runs across 3 countries and 500+ time series, with €15M+ annual savings in the first market's signed-off benefit case. Ships as packaged, unit-tested Python with CI on Databricks.
- Built a greenfield warehouse simulation to evaluate slotting changes counterfactually, modelling picker routes with shortest-path and travelling-salesman algorithms; handed over as the foundation a follow-up team built on
- Ran an internal talk series for the 40+-person data science community: probabilistic programming in PyMC, hierarchical forecast reconciliation, Prophet, Ruff, warehouse slotting optimization, and how our custom forecasting model works
Senior Data Scientist
Oct 2023 – Aug 2024Assortment Optimization
- Built a causal inference framework (S-learner) for assortment optimization, automating product delisting
- Mentored junior and mid-level data scientists
Senior Data Scientist
Jan 2022 – Sep 2023Product Recommender
- Built two real-time recommender systems (two-tower, TensorFlow) on billions of transactional rows, deployed on GCP / Vertex AI. A/B testing confirmed a conversion-rate lift; the business credited €10M+ in annual revenue once rollout completed after I moved on.
- Led technical hiring: screened 20+ candidates, co-led 5 interviews, onboarded 3 hires
- Mentored 3 data scientists and led upskilling events for a team of 20: recommender systems, word2vec, cost-sensitive learning
Senior Data Scientist
- Developed and open-sourced mamimo, a marketing mix modeling package (74 GitHub stars); results directly influenced marketing budget allocation
- Owned 3 analytical workstreams end-to-end: forecasting (Prophet, LightGBM, QlikSense dashboards), MMM, and causal inference
- Translated the model for budget-holders with no ML background, e.g. explaining overfitting as "memorizing the answers for a test, then facing a completely different exam"
- Mentored 2 junior data scientists
Data Scientist
- Predicted TV ad view-through rates for media clients: modeled what share of an advertising block audiences would watch
- Joined datasets without a common identifier using statistical record linkage (Nearest Neighbor matching)
- Built the data backend for a competitive benchmarking dashboard so a client could compare performance against direct competitors
Trainee Risk
- Automated processes in Risk and Cyber Security (Python, SQL, VBA), eliminating 2 FTE-equivalent manual workloads
Freelance
Guest Instructor & Open-Source Contributor
- Design and deliver the official sktime curriculum across 3 cohorts (15 professionals)
- Active open-source contributor: proposed improvements to StackingForecaster, pipeline caching, and GroupByTransformer
Consulting Data Scientist (Part-Time)
- Part-time advisory and backend engineering (PyMC-Marketing) for a SaaS analytics platform; engagement continued under ScanmarQED after its acquisition of MMMLabs.ai (Feb 2026)
- Work in direct, ongoing collaboration with the business lead: he brings the business need, we shape the solution together
Technical Author
- 90+ articles on data science and ML topics, with 1M+ total reads across platforms
Skills
Areas of expertise
Leadership & Vision
Setting direction, growing people
Bridging research and product: defining technical strategy, mentoring engineers, and making complex ideas land with any audience.
ML & Mathematics
Reasoning from first principles
Probabilistic thinking, Bayesian inference, and theoretical foundations applied to real forecasting and decision problems.
Engineering
Building reliable systems
Turning research-grade ideas into production-ready, maintainable systems on modern cloud infrastructure.
Community
Open source, talks, and volunteering
Open Source
scikit-lego
Regression models for zero-heavy, imbalanced, and uncertain count data.
PyMC-Marketing
Channel contribution attribution charts for marketing mix models.
Invited Talks
Introduction to Embedding-Based Recommender Systems
- A 45-minute talk on collaborative filtering via matrix factorization, building a recommender system from scratch in PyTorch
Introduction to sktime
- Data science meetup talk covering sktime's forecasting API and pipeline design for practitioners
How we recommend alternatives
- Talk on recommending alternative products when preferred items are unavailable, so customers can keep shopping
A/B Testing at Metro
- General introduction to A/B testing concepts and experiment design for a business-focused audience
Volunteer
Moderator, matheraum.de
- Supported school and university students with mathematical questions over 12 years
Data Science Mentor
- Guided aspiring data scientists through an end-to-end project predicting used car prices
Writing
Technical articles
Publications
Research papers
My PhD was about attacking a learning problem deliberately designed to be unsolvable. LPN (Learning Parity with Noise) is a machine learning problem engineered to resist efficient algorithms, which is precisely what makes it useful for cryptography. Building dedicated algorithms with precise runtime, memory, and success-probability guarantees developed the tools I reach for in production ML every day: probabilistic reasoning, algorithmic thinking, and comfort with problems that resist easy answers.
A Faster Algorithm for Finding Closest Pairs in Hamming Metric
- Presents a faster randomized algorithm for finding the closest pair of vectors under Hamming distance, with improved complexity for small dimensions relative to dataset size
Time-Memory Trade-offs for the Learning Parity with Noise Problem
- Investigates algorithms that trade memory for runtime when solving the LPN problem, a hard problem underlying several post-quantum cryptographic schemes
Dissection-BKW
- Applies dissection techniques to the BKW algorithm for solving LPN, improving the memory-runtime tradeoff
LPN Decoded
- A unified framework for analysing and comparing LPN solvers, introducing the first quantum algorithm for LPN and demonstrating practical attacks on medium-parameter instances
Education
Academic background
PhD, Mathematics (Cryptography)
- Dissertation: Time-Memory Trade-offs for the Learning Parity with Noise Problem, focusing on memory-efficient algorithms for the cryptographically relevant LPN problem
- 3 peer-reviewed publications (CRYPTO 2017, CRYPTO 2018, FSTTCS 2021)
- Taught cryptography, number theory, discrete mathematics; supervised 5 theses
MSc, Mathematics · Minor: Computer Science
- Grade: 1.2 (top 5%)
Japanese Language & Culture
- Japanese language: A1 → B1
BSc, Mathematics · Minor: Computer Science
- Grade: 1.0 (top 5%)
Recognition
Certifications
Languages
Projects
Open demos
parlascanned
How Bundestag members vote, speak, and earn
Interactive transparency dashboard for the German parliament, built on public disclosure data. Explore voting similarity across 630 MPs, track legislative initiatives, analyze debate topics with NLP, and browse side income by party.
Launch demo →Contact
Get in touch
Open to Staff and Principal level data science roles in NRW or remote within Germany, and to consulting, teaching, and speaking. Based near Krefeld, hybrid-friendly. Feel free to reach out about anything: articles, open-source work, career questions, or just to say hello. I read everything.