1. business related
2. there is some sort of association between outcomes/predictor and the variable.
- if both are continuous vars, correlation works
- if both are categorical vars, chi square test works
- if one categorical one continuous vars, t test works
3. toss all those vars that are business relevant or with some kind of assoication with outcomes/predictors into a model, such as linear, logistic, or some other regression which might be detemined by the industry/business, and then determine if the difference is still statistical signficant. We can tell that by checking the p-Value assoicated with that specifc regression coefficient.