Build a propensity model

1. Build a team of domain experts that includes email marketers, conversion optimizers, data scientists, finance experts, and CRM specialists.

Your team should include anyone with relevant business acumen.

2. Select which features, independent and dependent variables, to investigate in your model.

For example, your dependent variable may be lead-to-customer conversion while your independent variables may be product milestones, app and theme downloads, demographics, or buying history. There are mathematical and experience-based methods for choosing variables to test.

3. Work with your team to decide whether you want to interpret the regression coefficients or not.

Statistically, features that are less relevant, will have regression coefficients closer to zero.

4. Determine if linear or logistic regression is a more appropriate probabilistic model for your analysis, based on the features you selected.

In linear regression, the outcome is continuous, meaning it can have an infinite number of potential values. It’s ideal for weight and number of hours. In logistic regression, the outcome has a limited number of potential values. It’s ideal for yes/no and 1st/2nd/3rd.

5. Train your propensity model using a test data set, prior to calculating your propensity scores.

You can train your model on 50% historical data, and test it on the remaining 50%.

6. Experiment to verify the accuracy of your propensity model scores.

This allows you to validate the accuracy of your future propensity scores.

7. Combine your propensity model results with expert knowledge from internal domain experts to run smarter, more targeted experiments, and leverage future outcomes.

For example, optimizers can leverage the propensity to unsubscribe, by reducing the frequency of emails, or sending a special offer letter to reinforce the value of remaining a subscriber.