Introducing Meridian MMM

Today, we're releasing Meridian MMM, our implementation of Google's open-source Marketing Mix Modeling (MMM) framework

Meridian MMM marks our first deployment of a Bayesian modeling approach, enabling measurement capabilities beyond those achievable with traditional frequentist or evolutionary modeling methods, such as Robyn.

Our Approach to Bayesian Modeling

What is Bayesian modeling? Instead of only looking at historical data patterns (like traditional MMM tools such as Robyn), Bayesian modeling lets you incorporate what you already know works.

How we made it simple: When you run incrementality tests—like a Facebook lift study or YouTube brand lift—you can feed those real results directly into the model. The model then uses this proven data as a "starting point" and adjusts everything else around it. This means less guesswork and more accuracy.

You Should Consider Meridian When:

• You have more than 18 months of media spend and sales data

• You have reach and frequency data for key specific channels (like YouTube, Facebook, TV)

• You want a multi-market marketing mix model across different countries or regions

• You already have incrementality test results for your biggest channels

Key Features That Matter

Incrementality Calibration

Feed your lift test results directly into the model. If your Meta campaign delivered €50K in incremental sales over 30 days, the model treats this as fact and builds everything else around it. No more wondering if your MMM results match your test results.

Reach & Frequency Integration

For channels like YouTube, the model considers actual reach and frequency data instead of just impressions. This gives you better diminishing returns curves and more realistic forecasts for scale-up scenarios.

Multi-Market Modeling

Train models for multiple countries simultaneously. Faster setup, consistent methodology across markets, and shared learnings between similar regions. Note: Results aren't simply additive across markets, but you get better insights for each individual market.

Organic Search Control (GQV)

Use Google search volume data to separate organic demand shifts from your paid search effectiveness. Helps you understand what's driving baseline demand versus what your ads are actually contributing.

Why This Matters for Your Daily Work

Consistency: Models built from these additional data point, increase the causal estimation accuracy of your Marketing mix 

Less Manual Work: Automated validation checks catch issues early. Less time spent with uncertain measurement, you know what you need to do to make the model robust, and then start optimizing.

Better Measurement: The model continuously validates its measurement against new data as it comes in, flagging when channels become unstable.

Real-World Accuracy: By incorporating your actual test results, forecasts become more reliable for budget planning and scenario modeling.

Robustness & Validation

Our implementation of Meridian automatically runs accuracy tests using rolling time windows. It trains the model multiple times on different date ranges and compares results. When it spots inconsistencies, it flags which channels need more incrementality testing to improve accuracy.

Think of it as continuous quality control for your MMM, ensuring your measurements remain trustworthy as market conditions change.

Getting Started Without Perfect Data

The Reality: Ideally, you'd have incrementality test results for every single channel before building your Meridian model. In practice, most businesses only have lift tests for 1-2 channels—if any at all.

Our Solution: Instead of requiring channel-by-channel test results, you can start with a strategic prior. Simply estimate your overall media contribution to total revenue and your confidence level. Cassandra then intelligently distributes this across individual channels based on your spend patterns and starts training the model.

This approach lets you get started immediately while the model learns from your data. As you run more incrementality tests over time, you can feed those specific results back into the model for even greater accuracy.

Do you want to try it?

Our Modelers are ready to show you how you can start measuring incremental sales using Meridian.

Click here to book a Demo