Phase 4

Modeling

Use patterns in the prepared data to make a useful prediction.

What this phase actually is

Modeling is the phase most people picture first, but it only works because the earlier phases made the question and data usable.

A model learns a relationship between input signals and an outcome. In practice, the team usually compares simple baselines, interpretable models, and more flexible approaches before choosing what fits the decision.

The useful output is not “an algorithm.” It is a scoring method with known behaviour, known limits, and a reason to believe it can improve the real decision.

How this looks at Bertelsmann

Try it

Mini Churn Scorer

Illustrative formula

Days since last login 18 Watch hours per week 3.5 Support tickets 1 Plan tier Basic

Illustrative churn probability 48%

This is a transparent hand-crafted formula. It shows how signals push a score; it is not trained on real subscriber data.

Pitfalls

Treating a higher score as an explanation instead of a prediction.
Choosing a complex model before a simple baseline has been tested.
Forgetting that the model only sees what the data preparation gave it.