Saurav Mukherjee, a Senior Manager at Cognizant had this to say:
I was first introduced to the idea of Causal Inference at NYU Stern.
There is clear business benefit in thinking in terms of 'Why' in addition to asking 'How Much [ regression]', 'Who [ Logit / probit]' , 'Which [ Association ]' etc. I don't see organizations embracing the Causal Modelling. They are still stuck with Predictive. The reason could be that
- Causal is not definitive. Meaning we will never have a point estimate
- Requires strong domain knowledge
- Most organizations have not yet fully implemented the predictive aspect.
Let me give an example:
Estimated Stock Price can be loosely written as = Average Stock Price *d(t) + Volatility * d(z) based on Geometric Brownian Motion theory.
Now if we don't peel the onion [ in this case Volatility ] and understand the underlying causes we will end up with spurious correlations and ultimately incorrect predictions.
Understanding Causal Modelling can help us in identifying the factors in terms of significance and including the confounding and thereby improving the prediction model.
Here is a book on causal modelling that i found good [ extremely fat ]. You can pick and choose your area.
https://cdn1.sph.harvard.edu/wp-content/uploads/sites/1268/2019/10/ci_hernanrobins_1oct19.pdf