Is there any guide or experience for setting the treatment in observation data？

[UTimeStrange]

A 0-1 value treatment column is needed for a Casual Inference task.
But in our observation data, there is no such strict 0 - 1 value column can be used as treatment directly. In fact, most features are consecutive or one hot value with multiple distinct values. If we wonder which feature uplift the KPI(y) most, we have to chose a feature and make it split into 2 values(higher or lower than a chosen line for consecutive). It's not that convinient to find such "line"for spliting.
So, is there any guide or experience for setting the treatment in observation data？