MMM Upgrade from PyMC3 to PyMC-Marketing with HSGP

Fabletics

Retail & E-Commerce
PyMC-MarketingJAXMLflowCausalPy

The Challenge

Fabletics had a working marketing mix model — but the person who built it had left the company. The remaining team understood the business questions but needed expert guidance to modernize the model's architecture. They wanted to upgrade to a current framework, capture how marketing effectiveness shifts over time rather than treating it as static, and lay groundwork for future capabilities like experiment-based validation of their media spend.

Our Approach

We worked alongside the Fabletics team in a coaching model — they wrote the code, we guided the decisions. The core upgrade moved their marketing mix model to a modern framework and introduced time-varying parameters so the model could reflect how channel effectiveness evolves quarter to quarter rather than assuming it stays fixed. Along the way, we diagnosed a significant performance bottleneck in the model's media transformation step and implemented a fix that restored practical runtimes. We also mapped out a phased plan for production deployment and identified geographic experimentation as a natural next step for validating model-derived recommendations with real-world tests.

Results

Within a few months the upgraded model was producing clean, credible estimates. The internal team gained enough confidence and skill to own the model going forward.

One particularly satisfying outcome: the team lead driving the project was promoted during the engagement, recognized for the modeling work she'd taken on.

“PyMC Labs was instrumental in helping us implement time-varying coefficients to better capture seasonality in our marketing mix models, significantly improving both model accuracy and interpretability. The team was collaborative, insightful, and consistently supportive throughout the engagement.”

Kate Hirth , Senior Data Scientist, Fabletics

PyMC Labs Team

  • Niall Oulton
  • Bill Engels
  • Juan Orduz
  • Will Dean
  • Luciano
  • Jesse Grabowski

More Work in Retail & E-Commerce

HelloFresh

Rebuilt HelloFresh's marketing measurement and testing infrastructure, cutting model runtimes from 20 minutes to 2 and overnight test pipelines from 6 hours to 6 minutes across 15+ global markets.

MMMGaussian ProcessesHSGPA/B Testing

Swarovski

Improved prediction accuracy by 20% for Swarovski's marketing mix model by letting the baseline sales level shift naturally over time instead of holding it fixed.

MMMHSGPTime-Varying InterceptAdstockSaturation

L.L. Bean

Helped L.L. Bean's team build their own marketing mix model spanning 50 US markets with catalog attribution, replacing a vendor model they couldn't inspect or modify.

MMMHierarchical Bayesian MMMHSGPGeo TestingSynthetic Control

Let's Chat, We Respond Fast

Tell us about your problem. We typically respond within 24 hours.

Schedule a Consultation