Attribution is getting harder as privacy rules tighten. Open-source marketing mix modeling tools are now more accessible, with AI making them usable for non-experts. Marketers face new opportunities and risks.
As privacy regulations disrupt traditional attribution, open-source marketing mix modeling (OS-MMM) is emerging as a practical alternative for marketers and publishers seeking reliable measurement. Tools like Google's Meridian, Meta's Robyn, and PyMC Marketing are now widely available, offering transparent code and reusable frameworks that were once limited to data scientists and consultants. The rise of agentic AI further lowers the barrier, allowing even those with minimal modeling experience to run complex analyses with simple prompts.
Julian Runge, assistant professor of marketing at Northwestern University and co-author of the first academic paper on open-source measurement, said the shift is driven by privacy changes such as GDPR, COPPA, and Apple's IDFA deprecation. These developments have made attribution less dependable, pushing both platforms and advertisers to explore new solutions. Runge noted that while large brands have used marketing mix modeling for years, digital-first advertisers are only now adopting these methods as open-source tools become more accessible.
Each OS-MMM package offers distinct advantages. Robyn, released by Meta in 2021, is considered the most user-friendly and uses a frequentist approach, focusing on observed data. Google's Meridian and PyMC Marketing both employ Bayesian hierarchical frameworks, which incorporate prior knowledge and update as new data arrives. PyMC is the most customizable but requires greater expertise. Despite their origins in major platforms, Runge said there is no evidence of built-in bias favoring specific ad channels, though each tool reflects the orientation of its creator-Meridian toward view-based inputs and Robyn toward spend and sales.
AI agents like Claude, ChatGPT, and Gemini can now automate much of the OS-MMM process. Users can instruct these agents to install packages, load data, and estimate models, making advanced analytics accessible without deep technical skills. Because marketing mix modeling typically involves small datasets, the process is manageable even for those without premium AI subscriptions. However, Runge cautioned that while AI can execute instructions, human oversight remains essential to interpret results and avoid overconfidence in outputs that may not be fully vetted.
The democratization of OS-MMM means marketers can use large language models to clarify concepts and refine questions before consulting data scientists. Yet, the risk of misinterpretation or overreliance on AI-generated analyses persists. Runge emphasized that while AI can surface novel insights, it is up to human experts to judge their validity and practical value.
Looking ahead, Runge expects that marketing measurement will continue to rely on a combination of marketing mix modeling, experiments, and attribution-what Meta calls the "suite of truth." MMM provides a strategic overview, experiments offer ground truth for calibration, and attribution supports daily tactical decisions. Multi-touch attribution still has a role but requires careful scrutiny, as different models can yield widely varying results. Practitioners are advised to cross-check findings through experiments to ensure reliability.
For those interested in how AI is reshaping advertising and measurement, recent developments in conversational ad platforms are also worth noting, such as the expansion of ChatGPT ads for SMBs, as discussed in this related report.
