The pre-modeling process is a crucial first step in building a successful Marketing Mix Model (MMM). This stage involves gathering, cleaning, and organizing your data to ensure it's ready for modeling. A thorough pre-modeling phase helps to build a reliable model. This section provides a guide to the key steps in the pre-modeling process.
| Page | Description |
|---|---|
| Collect and organize your data | Learn how to gather and arrange your data for Meridian. This guide covers the essential data you'll need, like media spend, KPIs, and control variables. It also discusses recommendations for how granular your data should be in terms of geography, time, and media channels. |
| Amount of data needed | Find out how much data you need for your Marketing Mix Model. This page explains how the right amount of data can vary. It covers the different data requirements for national versus geo-level models and offers solutions for situations where you might not have enough data. |
| Geo-level modeling | Discover the advantages of using data broken down by geographic regions instead of national-level data. Using geo-level data can lead to more accurate and reliable model results by improving statistical power and reducing potential biases. This guide also offers tips on choosing your geographic areas and incorporating national data into a geo-level model. |
| Use MMM Data Platform | This guide shows you how to use the Google MMM Data Platform to source important data for your model, such as Google Search trends and YouTube metrics. You'll learn how to manage access, set up data delivery, and make data requests. It also breaks down the structure of the data you'll receive from various Google advertising platforms. |
| Perform an exploratory data analysis | Learn how to perform an exploratory data analysis (EDA), a critical step before building your model. This process helps you find and fix issues in your data, ensuring your model is built on a solid foundation. You'll learn how to check for missing or inaccurate data, correct errors, and look for strong correlations between variables that could affect your model's performance. |