Chapter 2: Validate the choice of model

The best practice to validate the choice of a model is to use the incrementality measurement tools offered by advertising platforms, notably those of Meta and Google.

Whether it is a Single-Touch, Multi-Touch, or Marketing Mix Modeling model, incrementality tests will help validate channel contribution metrics.

Since attribution models work primarily on conversion goals, we will focus here on conversion incrementality testing (as opposed to brand awareness incrementality testing).


Step-by-step procedure for incrementality testing on Meta and Google Ads platforms



Step 1: Define objectives and hypotheses

  • Set clear goals: Identify what you want to measure with the lift test, such as increased sales or increased revenue. Clearly describe the business question you want to answer .
  • Formulate hypotheses: Develop a hypothesis related to your goals. For example, “Running ads on Meta will increase sales by 10% compared to no ads” .


Step 2: Choose the right type of incrementality test

  • Meta Ads Platform:
  • Use conversion testing to measure the increase in conversions or sales caused by ads.
  • Google Ads Platform:
  • Use conversion leverage measurement via randomized control groups to measure direct impact on conversion.


Step 3: Plan and design the incrementality test

  • Define the test duration and budget:
  • Make sure the test runs long enough to collect enough data, usually at least two weeks , but this duration can vary depending on traffic volume and conversion rates.
  • Allocate sufficient budget to generate meaningful data, ensuring that test and control groups are adequately represented.


Step 4: Configure incrementality testing on each platform

  • On Meta Ads:
  • Go to the Experiments tool in Meta Ads Manager.
  • Choose the appropriate test type.
  • Define your primary metric (e.g. conversion rate or sales volume) and set up your test and control groups.
  • Ensure proper implementation of Meta Pixel, Conversions API or other tracking mechanisms to capture data accurately.
  • On Google Ads:
  • Use Drafts & Experiments to set up incrementality testing.
  • Set test parameters (e.g. conversions) and select a split method (50/50 split is common).


Step 5: Check and validate data quality

  • Check data accuracy: Make sure the data is consistent. Look for discrepancies or missing data points.
  • Monitoring of key indicators of Real-time performance: Monitor key performance indicators during testing to ensure it is running as planned and data collection is smooth.


Step 6: Analyze the results and measure the incremental lift

  • Measure the incremental lift:
  • For Meta, the platform automatically calculates the uplift based on the difference in results between the test group and the control group.
  • For Google, use the Conversion Lift report to compare results from the test and control groups to ensure statistical significance.
  • Use Meta's Confidence Levels: Look at the confidence levels provided by Meta (typically this is at least a 90% confidence level).


Step 7: Interpret the results and make decisions

  • Compare with hypotheses: Check whether the test results confirm or refute your initial hypothesis.

Choose the attribution models whose Meta and Google contributory weights will be the most consistent with the results announced by the incrementality tests!


Other good practices

  • Ensure sufficient power: Before performing the test, perform power calculations to ensure that the test is powerful enough to detect significant lift.
  • Minimize contamination: Avoid overlapping test and control groups in terms of media exposure or external factors that could confound results.
  • Regular Calibration: Regularly calibrate models and test configurations to account for changing market dynamics or platform updates.