To chart a clear course, MMM Pilot requires a specific set of data points. Think of this as your flight plan: the more accurate the information you provide, the smoother the journey will be.
You don’t need complex integrations. You simply need a single CSV file containing the following three categories of data.
Download a sample file: CSV Format, MS Excel Format
1. The Destination: Dependent Variable
This is the main metric you want to explain. It is the “output” of your business.
- What it is: Usually Total Revenue (Sales) or Total Conversions (number of orders/leads).
- Tip 1: Choose the metric that best reflects your actual business growth. This is the number the model will try to predict.
- Tip 2: Try to avoid GA4 data if possible. See why.
2. The Fuel: Paid Media
These are your marketing efforts—the “inputs” you pay for to drive the destination.
- Spend: The amount of money invested in each channel (e.g., Facebook, Google, TV).
- Impressions (Exposure): While spend tells us the cost, impressions tell us how many people actually saw the ad.
- Why both? MMM Pilot supports “paired” variables. Using both spend and impressions helps the model understand if your ads are becoming more expensive (CPM increases) or if they are simply reaching fewer people.
3. The Weather: Contextual Variables
Your marketing doesn’t exist in a vacuum. External factors influence your sales just as much as your ads do. The model needs to know about the “weather” conditions surrounding your business.
- Seasonality: Natural peaks and valleys (e.g., Christmas spikes or summer slumps).
- Holidays: Specific events like Black Friday or Cyber Monday.
- Economics/Competitors: If you have data on competitor price drops or major economic shifts, including them helps the model distinguish between a bad ad campaign and a bad market week.
Why Detail (Granularity) Matters
The model searches for patterns over time.
- Consistency: Data should be organized in consistent time intervals—usually weekly or daily.
- The “Why”: Granular data allows the model to see cause and effect more clearly. If you only provide monthly data, it is very hard to see if a specific ad campaign caused a spike in sales. Weekly data gives the model more “clues” to solve the puzzle.