Open-source marketing mix modeling tools gain traction as AI agents lower entry barriers

August 18, 2026 · AdExchanger

Privacy changes including GDPR, COPPA and Apple's deprecation of the IDFA have made traditional attribution unreliable, pushing marketers toward open-source marketing mix modeling, or OS-MMM, according to Julian Runge, an assistant professor of marketing at Northwestern University and co-author of the first academic paper on open-source measurement. Google built Meridian, Meta built Robyn and the PyMC-Marketing library offers a Bayesian alternative, and all three publish their code openly for inspection. Runge told AdExchanger that agentic AI now lets anyone with natural-language skills prompt tools such as Claude, ChatGPT or Gemini to pull a package, load campaign data and estimate a model, calling the drop in required expertise "an increase of infinity, in a way."

Teams that once needed a dedicated data scientist to build a mix model can generate a first pass by prompting an AI agent directly, then bring refined questions to a specialist for review. Runge warned that skipping supervision invites overconfidence, since an agent can produce a plausible but flawed analysis that looks finished. He expects marketing mix modeling, controlled experiments and attribution, Meta's 'suite of truth,' to remain the standard measurement stack, with multi-touch attribution results still needing validation against experiments given that different models can diverge by orders of magnitude.