Tldr: is fitting a Logit model on observational data and explaining the coefficients in the causal manner a terrible idea assuming the X variables aren't too crazy.

So, for context. The goal is to understand what causes store promotions work - which are driven in big part by how the executions go: perfect execution - great result vice versa. The goal is not really to tease out the influence of macro drivers on demand or micro customer level factors either. It's mostly to pin point operational breaking points and quantity them.

To learn about the operational breaking points we don't need any causal model just pure data exploration is good enough. But manager has been pushing to deliver some kind of model based analysis to provide what causes an initiative produce result. He comes from non DS background and from what I gather he just want to have some brownies points saying we ran some DS models to come up with numbers.

So with that in the background, would that be a too terrible idea to just run plain LR on the observational data and explain the coeffs like they mean causality while they don't actually and the model is riddled with tons of biases that we can point out?