Causal AI at KDD 2024: Why companies that don't jump on the causal train now will have a harder time competing in 2025 and beyond | by Aleksander Molak | September 2024

Developing causal experience is a process, not an event

doAusal modeling is a general term for a wide range of methods that allow us to model the effects of our actions on the world.

Causal models differ from traditional machine learning models in several ways.

The most important distinction between them arises from the fact that the information contained in the observational data used to train traditional machine learning machinery is generally insufficient to consistently model the effects of our actions.

The result?

Using traditional machine learning methods to model the results of our actions leads to biased decisions most of the time.

A good example here is using a regression model trained on historical data to determine your marketing mix.

Other?

Use XGBoost trained on historical observations to predict churn probability and send a campaign if the predicted churn probability is greater than some threshold.