Meridian 1.3.0: New quality checks, geo-level optimizations, and more!

[kmm022]

New update (as of 1.3.0) – Meridian now offers automated quality checks and geo-level optimizations, empowering you to build better models and run optimizations aligned to how you make decisions.

1. Data Quality Checks – Before you run your model, Meridian will automatically perform data checks and flag critical issues that would lead to extreme modeling errors. If any critical issue is found in the data, you’ll see an error detailing the issue and how to fix it before moving forward. Once you’ve resolved any data errors, you’re ready to move onto configuring your model. These pre-modeling checks will save you time and improve your model by alerting you to resolve critical problems before investing time and resources into modeling.

2. Model Quality Module – Meridian’s new post-modeling checks help you ensure your model’s quality before using it to inform critical business decisions. You’ll learn how to diagnose common issues, including negative baselines, Bayesian p-value, prior-posterior shift, ROI consistency, and more. These checks will make it easier to quickly troubleshoot potential problems, better understand how your data informed your model, and build confidence that the model's outputs are trustworthy and realistic.

3. Geo-level optimization – Meridian enables you to analyze and optimize your model for specific geographical region(s), empowering you with granular insights aligned to how you make business decisions. With the “selected_geos” and “selected_times” arguments, you can calculate metrics (like ROI or incremental outcome), generate visualizations, and perform budget optimization based on subsets of your input data.

4. Serde – This is Meridian’s new Protobuf-based framework that allows you to easily store and load your trained models as single, portable files. Serde provides enhanced security, easier debugging, and reliable backwards compatibility for your models.

We’ve also added the ability to customize the currency symbol in your optimization output and enabled fine-tuning capabilities in Automatic Knot Selection to be used in complex edge cases when a high number of knots are suggested but data is limited.