When you create a custom prediction of Binomial classification type, you may notice that many customers have the same score or that scores are grouped into just a few values.
This behavior is usually expected and comes from how tree‑based models work.
How prediction scores are generated
Custom binomial predictions are based on decision trees:
-
You define:
Target filter – what you want to predict.
Eligible filter – which customers you want to predict for, for example, active buyers.
Features – events and properties that should influence the prediction.
-
During training, the model:
Splits customers into groups based on their behavior, such as events, attributes, and aggregates.
Creates a branch for each final group.
-
For each branch, the model calculates:
In this group, X% of customers reached the target.
-
When you use the prediction:
Every customer passes through the same tree.
Based on their current data, they land in one branch, or behavior group.
They get the same probability as all other customers in that branch.
Outcome
As a result, the model does not predict a unique value for each customer. Instead, it predicts a probability for each group of similar customers.
If the tree has only a few branches, it can return only a few distinct probability values. This is why you may see many customers sharing the same score, or scores clustered into a small set of values.