Training a Predictive Intelligence model using external data

[Mahesh_Krishnan]

Hello All:

We have built a scoped app with an initial focus on creating interactions from emails and using AWA to route the interactions to the correct queue. As it is a greenfield implementation and we do not have historical email interactions in ServiceNow I wish to know how we might be able to use data from an external system as input to train a PI model in ServiceNow. In an external ticketing system we have data containing specific product topics for each ticket and the call history that contains the email content. My thought is if we can upload that data into ServiceNow and use a Supervised Classification model and provide the call history data as input and the Product Topic as output to train the model we can call the model in our inbound flow, pass it the email body and get the appropriate Product Topic to then associate it with the Interaction. This would enable us to then associate a Product Topic to a Skill as implement Skills-Based routing in our AWA logic.

[SoureshD6782072]

Yes, your proposed approach is valid: you can use historical ticket/email data from an external system to train a ServiceNow Predictive Intelligence Supervised Classification model, provided that the data is imported into ServiceNow in a clean, structured, and labelled format.
Predictive Intelligence classification is designed for scenarios where you have historical input data and a known target/output value, such as predicting a category, product, assignment group, or topic from text fields. ServiceNow documentation describes classification as a framework for predicting field values from historical records, and trained solutions can be invoked by applications through an API. [servicenow.com], [servicenow.com]
In your case:

Input: historical email body / subject / call history text
Output: Product Topic
Runtime use case: inbound email creates an Interaction, PI predicts Product Topic, then AWA routes based on Product Topic / Skill

That is a strong use case for Supervised Classification.

[Tanushree Maiti]

Hi @Mahesh_Krishnan 

ServiceNow Predictive Intelligence (PI) fully supports custom tables. You can train Classification, Similarity, Clustering, and Regression machine learning models on your own tables to auto-populate fields, detect duplicates, or route tasks.

Refer: 

Predictive Intelligence Made Simple – A Use Case to Help You Get Started

https://www.servicenow.com/community/developer-forum/script-include-not-returning-any-value-when-cal...

https://www.servicenow.com/docs/r/yokohama/conversational-interfaces/virtual-agent/tr-troubleshootin...