How to implement a predictive model?


Might be a very basic question, But just wanted to know. How the predictive models that are built, tested with some test data are implemented to the production data?

I previously worked on mainframe environment, where the programs run in batch mode in the nights and processes the data. How should we think about in data analytics? Is that something the R program that we write runs similarly in a batch mode at certain intervals?

Any sort of advice or pointers to certain links would be really helpful. I could not find one.



The production environment and settings would vary from company to company and situation to situation. It would also depend on the problems and the capabilities of the softwares already present in the organization.

So, in order to help, I am adding a couple of examples I know of:

  1. A cross-sell / up-sell campaign at a bank / telecom provider: In the setups I have seen, the modelling typically takes place outside the CRM. You run the model on customers and then upload the scores / segments into the CRM. The campaigns are then designed in the CRM. The model is usually run at a regular pre-defined frequency or the time when campaign has to be launched.

  2. Credit risk monitoring: Again, model gets built outside and the logic is fed into the decisioning system (typically known as credit policy). The output is typically 3 segments based on qauntiles / score - Accept, Reject, needs manual processing. Manual processing is the grey area, which might need a closer look. For example, limited credit history available or salary information not available.



Thanks a lot sir, thats making sense.