People could just define a target and let the system automatically select for them the list of system indicators that better predicts the target.

We would need to expand machine learning backends API for this.

This is more convenient but not 100% ideal either because the list of indicators will be selected based on the training data, which will be the data available at the time of creating the model. If there is enough data this should be ok, but if there is not much data the list of indicators that correlate well with the target may be different as there is more data in the site.