Follow Lauri in LinkedIn

Original LinkedIn post

💡 Next Gen Marketing Mix Models are calibrated with Experiments AND Attribution data

Over the years, we at Sellforte have been big fans of Meta's, and especially Igor Skokan's, pioneering work on driving and popularizing the concept of calibrating MMMs with incrementality tests to improve reliability of MMM results. Our own approach to model calibration has been heavily influenced and inspired by this work.

At the same time, we’ve been surprised too see how polarizing the idea of model calibration has been. ”Traditionalists” seem to be sceptical about the idea of incorporating anything else than modeling data into MMM, whereas the new wave of MMMs, which we call ”Next Gen MMMs” embrace the idea of using all available information to improve modeling results and their reliability. Next Gen MMMs combine

1) The power of the model to estimate marketing’s ROI, and

2) The richness of the information about marketing’s performance that exists outside the model.

Today, we're sharing more details on the model calibration approach in Sellforte’s Next Gen MMM solution. We build on the idea of calibrating MMMs with incrementality tests, and expand calibration data to include attribution data (from ad platforms, Google Analytics 4, and multi-touch attribution). We argue attribution data to be another important calibration dataset that contains prior information about marketing's ROI.

In our latest long-form blog post we will go through

🏢 Industry context: Different camps emerging around model calibration

🚀 Why are Next Gen MMMs using model calibration?

❓ How does model calibration work?

📐 What data do we use in model calibration?

⚡ Typical criticism against model calibration