Production MMM: Why Uncalibrated Models Fail and How Meridian 2.1.0 Solves It

Marketing measurement is fundamentally a causal inference problem constrained by observational data. The core challenge is observational confounding. Historical ad spend is rarely randomized; rather, ad pacing is strictly endogenous to expected demand. Advertisers increase budgets during high-conversion periods, creating a spurious correlation between capital allocation and baseline organic sales.

If you apply a naive Markov Chain Monte Carlo (MCMC) sampler to this structural data-generating process, it will systematically conflate endogenous spend pacing with baseline organic demand. The likelihood geometry cannot mathematically distinguish between the marketing causing the sales versus the sales expectations causing the marketing. This results in upwardly biased Return on Ad Spend (ROAS) parameters.

Over at Vector1 Research, I just published a complete architectural deep dive detailing how Google Meridian 2.1.0 escapes this endogeneity trap.

Meridian solves the identification problem through a Bayesian experiment calibration bridge. By formulating empirical treatment effects from matched-market geo-experiments as highly informative priors over the media effect parameters, the model is forced to allocate the remaining variance to the structural baselines and exogenous controls.

The article also unpacks the platform engineering required to scale this architecture. Meridian 2.1.0 shifts the computational heavy lifting entirely to JAX, relying on its High Level Optimizer (HLO) and Accelerated Linear Algebra (XLA) compiler. When evaluating hundreds of Designated Market Areas (DMAs), standard hierarchical architectures stall moving data matrices between host and accelerator memory. Meridian compiles the entire MCMC computational graph directly to the hardware. We examine exactly why executing this scale of Bayesian updating requires maintaining 64-bit precision to prevent the No-U-Turn Sampler (NUTS) from suffering divergent transitions during Hamiltonian leapfrog integration.

If you build production econometric systems, high-altitude summaries are useless. You need the structural equations, the memory mechanics, and the executable code.

The full article covers the mathematical proofs, the object-oriented partition serialization in the new API, production failure modes (including temporal aggregation bias and SUTVA violations), and an end-to-end Python code block for declarative model specification.

Read the full rigorous implementation on Substack. https://brianjamescurry.substack.com/

About the Author

Brian Curry is a Kansas City-based machine learning and AI strategist, marketing economist, and applied researcher. He leads Vector1, an independent research studio working on cognitive AI, causal inference, marketing economics, and knowledge engineering, with a focus on simulation-driven intelligence systems.

Over more than twenty years, Brian has led data science and analytics initiatives at enterprises ranging from $4B to $125B in revenue, including Koch Industries, Tractor Supply, Vail Resorts, Hallmark, Garmin, and AT&T. His work applies economics and causal inference to connect marketing investment to structural outcomes, operational decisions to economic value, and AI adoption to durable competitive advantage.

He is also the founder of KC AI Lab, a 1,500-member AI and ML education community in the Midwest. He lives in Kansas City and is an active part of the Kansas City community.

Research and Open-Source Projects

The modeling approach in this paper draws on PyCausalSim and Papilon.

Causal Inference and Economics

• PyCausalSim: Simulation-driven causal discovery and evaluation, using counterfactual scenarios to isolate structural drivers.
• Papilon: Complex-system simulation and optimization, with applications in media mix modeling and adaptive architectures.
• Daedalus: Causal analytics for money-flow mapping and macroeconomic tracking.
• The Autonomy Economy: A macroeconomic framework for the combined effects of rapid productivity growth and structural price deflation in an AI-automated economy.

Forecasting and Applied Modeling

• Regimecast: Time-series regime detection, structural breaks, and economic forecasting.
• CarryNet: Financial modeling for commodity carry trade analysis.
• Foodmesh-ml: Machine learning, optimization, and simulation for distributed food systems.

Cognitive AI and Semantic Architecture

• Memory-Node Encapsulation (MNE): A cognitive AI architecture and data structure for artificial episodic memory (NeoCortex-M).
• MeaningFlow: Semantic content modeling, embeddings, clustering, and coverage gap analysis.

Contact: brian@vector1.ai | GitHub @Bodhi8 | Substack

The text of this article may be shared with attribution under Creative Commons Attribution 4.0 International (CC BY 4.0). Code and models referenced here are available separately to paid subscribers.