Planning to enable Task Intelligence for ITSM in it in Prod.

SurenderReddy
Tera Contributor

4 hours ago

Hi All,

 

We have enabled Task Intelligence for ITSM in non‑prod and are planning to enable it in prod. Is it recommended to export/import the model via Update Sets, and will this also move non‑prod training or test data into production un-necessarily? Or is it best practice to configure and train Task Intelligence separately in Production?
0 Helpfuls
  • 134 Views
2 REPLIES 2

Ken tang
Tera Expert

2 hours ago

Hey, good question and a common one when people first get here.

 

Short answer: you won't accidentally move training data via update sets. Training data stays on the instance where training happens -- it's historical records in your database, not metadata. Update sets only capture the model configuration and definition, not the underlying data used to train it.

 

I'd go with exporting the trained model from non-prod and importing it into prod. ServiceNow actually supports this explicitly -- there's a built-in export/import path for Task Intelligence models specifically so you don't have to recreate from scratch.

- your non-prod model was trained on real production data (assuming you refreshed your instance), so the learned patterns should translate well
- retraining in prod means burning compute on your live instance and going through the whole assess-and-deploy cycle again when you already have a validated model sitting there
- training in prod also means your model is in a draft/training state while users are actively hitting incidents, which is just unnecessary noise

 

Once you're live in prod, plan to retrain periodically using fresh prod data. The model you bring over is a starting point, not set-and-forget.

 

I found this watch walks through the steps https://www.servicenow.com/docs/r/it-service-management/task-intelligence-for-itsm/using-task-intell...

0 Helpfuls

SurenderReddy
Tera Contributor
In response to Ken tang

4m ago

Hi @Ken tang 

Thanks for the response.

I have a follow‑up question to clarify the deployment approach. If we use the Export Model option in a non‑production instance and import it into the target (production) instance via Update Sets, my understanding is that we would still need to perform some manual prerequisite steps in production, such as:

  • Plugin installation
  • Plugin activation

Please confirm if this understanding is correct.

Additionally, I’d like to clarify the point you mentioned earlier regarding retraining the model in production. You noted that retraining in prod would involve consuming compute resources on the live instance and going through the full assess and deploy lifecycle again, even when a validated model already exists.

In that context, if we:

  • Refresh production data into a non‑production instance, and
  • Train and validate the Task Intelligence model using that data in non‑prod before exporting it,

would this approach allow us to avoid retraining the model again in production, and instead simply deploy the already trained model?

Please confirm whether this approach is supported and recommended from a Task Intelligence best‑practice perspective.

0 Helpfuls