Why start using uplift models for more efficient marketing campaigns

https://www.youtube.com/watch?v=2J9j7peWQgI 

Table Of Contents

1. Traditional Models for Marketing 

2. Why Uplift Modeling?

3. What is Uplift Modeling?

4. Algorithms in CausalML Package

5. Use case @ Uber

 

1. Typical Marketing Models

 

2. Building a model without a clear use is always dangerous.

-> Mismatch between model and business problem.

 

The common business goal is to maximize profit, in other words, increase revenue & save cost.

 

3. Missed Opportunities - Propensity Model

The high propensity customers here, so for those customers we actually can break them into three groups.

 

4. A/B Testing

-> Where A/B Testing falls short

 

Standard A/B Testing using ATE (Average Treatment Effect)

 

=> But We lost the opportunity actually to personalize the experience all the campaigns for individual customers.

 

5. Then How?

 

Uplift Modeling optimizes for incremental effect

Uplift Modeling enables personalized treatment

 

6. Uplift Modeling

Uplift model estimates heterogeneous treatment effects with ML algorithms.

 

Conditional average treatment effect: CATE = E [ Y | Intervention, X] - E[ Y | Intervention, X]

 

 

    a. Sure things & Lost causes - Will behave the same no matter what you do. Including them as a target in the model is okay, but will make our targeting inefficient.

    b. Sleeping dogs - These people are turned off by your intervention. Definitely don't include them, ideally you would even downrank them.

    c. Persuadable - This is the population you actually care about because they exhibit the ideal behavior because you intervened. Ideally you uprank them as much as possible.