Predictive intelligence deploy process

[Kohei Tominaga1]

Hi,

I am learning the Predictive intelligence solutions so that I can use the OOTB HR case auto categorization capability.

I understood that ServiceNow recommend to prepare the learning data by cloning Prod to non-Prod. However the case data sometimes include personal data which would not be permitted by our COE to have in non-Prod.

If model need to be created first in non-Prod, I need to check tha massive data and mask the personal data. Or, is it also recommended to train model in Prod directly?

Could you please share your experience?

[Lener Pacania1]

Hi Kohei, if you must train in production make sure you turn off the sn_hr_core.case_auto_categorization system property (see the docs).  If you don't the trained model will go active and and start predicting and overwriting the HR assignment group in production, which may be undesirable.   That's why we recommend UAT in to a cloned sub-prod.  -Lener

[Kohei Tominaga1]

Hi, @Lener Pacania1 
Thank you for your comment.
Yes, I agree that we should do UAT in sub-prod so that operating team would understand how it works.

So, I have to think of preparing data set with personal data excluded... Do you have any project experience cleaning and masking data set? 

[Lener Pacania1]

I'll ask the other members of the community to chime in.  One approach is to use ServiceNow Vault which will anonymize/mask data so that developers can use sensitive data in non-prod.    There are probably other ways of doing it.