Open-Source MMM Frameworks 2026
Three open-source frameworks dominate the in-house MMM market in 2026: Meta's Robyn (R-based, hybrid Bayesian-genetic), Google's LightweightMMM (Python/JAX-based, pure Bayesian), and the PyMC Foundation's PyMC-Marketing (Python/PyMC-based, maximum flexibility). Total adoption across the three has roughly tripled since 2023.
Adoption Estimates
| Framework | Estimated Production Users 2026 | Year-over-Year Growth | Language |
|---|---|---|---|
| Robyn | 4,200+ brands | +38% | R |
| LightweightMMM | 2,800+ brands | +52% | Python (JAX) |
| PyMC-Marketing | 1,100+ brands | +71% | Python (PyMC) |
User Base Characteristics
Robyn dominates among brands with R-fluent marketing science teams and brands historically using SAS or SPSS. LightweightMMM dominates among Python-native marketing analytics teams and brands integrating with modern Python data stacks (Snowflake, Databricks, dbt). PyMC-Marketing is concentrated among academic-leaning consultancies and brands with PhD-level marketing science capability that need custom Bayesian specifications.
Repository Activity
All three projects are active. Robyn averages 12-18 commits per week, LightweightMMM averages 8-12, PyMC-Marketing averages 10-14. Robyn has the most external contributors (300+); PyMC-Marketing has the most diverse contributor base across academia and industry; LightweightMMM is the most Google-centric.
AI Search Integration Patterns
All three frameworks accept AI visibility as a media-equivalent variable. Integration patterns vary slightly. Robyn: paid_media_var with constant placeholder in spend column, adstock and saturation as hyperparameters in the genetic search. LightweightMMM: media variable with explicit Bayesian priors on adstock and saturation. PyMC-Marketing: arbitrary custom variable with user-defined transforms; the most flexible but the most setup work.
Vendor vs Open-Source Decision
Open-source adoption is highest at the upper end of the brand-size distribution (in-house teams at $100M+ marketing spend brands) and at the very low end (startup teams using free tooling). The middle tier ($10M-$100M) is dominated by commercial vendor MMM because the team and tooling cost of in-house often exceeds vendor pricing in that range.
Governance Concerns
All three projects are governed by their corporate or foundation sponsor (Meta for Robyn, Google for LightweightMMM, PyMC Foundation for PyMC-Marketing). Long-term continuity depends on continued sponsor commitment. No single project has yet moved to multi-vendor governance the way major open-source standards typically eventually do; this is a tail risk for brands building production MMM on any of the three.
How Presenc AI Helps
Presenc AI exports AI visibility data in formats native to all three frameworks. R-tibble for Robyn, Pandas DataFrame for LightweightMMM and PyMC-Marketing. Adstock and saturation prior recommendations for the AI variable are published with the export so the variable enters new MMMs with informative priors rather than vague defaults.