on - edited on by
Steph Morillo
Overview
Task Intelligence for ITSM uses a guided easy to use approach to deploy machine learning AI models that can predict desired business outcomes to help streamline ITSM workflows and processes.
Important Quick Links on Task Intelligence for ITSM:
- Task Intelligence for ITSM - Detailed Implementation Guide on NowCreate (login required)
- AI Academy YouTube video on Task Intelligence for ITSM
- Explore Task Intelligence for ITSM
- Managing Task Intelligence for ITSM models
- Export an Incident prediction model
- Monitoring and Analytics for Task Intelligence for ITSM models
Task Intelligence is available for ITSM in Utah-Patch 5, Vancouver and later releases.
|
Plugins and Entitlement |
ITSM PRO subscription is required.
Plugins needed (latest available):
Additional Plugins needed (latest available):
The following plugins needed for Language Support:
|
|
Roles |
Find out more about roles in the docs. |
|
Self-Hosted and regulated market |
Task Intelligence is available for most regulated market data centers. Task Intelligence is not yet available for self-hosted customers. |
Post Plugin Implementation Steps for Task Intelligence for ITSM:
Configure the Task Intelligence for ITSM model from Admin Console:
- Navigate to Task Intelligence for ITSM > Setup to open the Task Intelligence Admin Console
- Click on ITSM OOB TI Solution OOB shipped model to open Train your model tab in Admin Console.
- If needed adjust the Conditions filter here, which determine the incident data that the model will be trained on.
- Observe that the OOB shipped model comes pre-configured to predict on Assignment Group, Configuration Item, and Service as the Output fields fully configured for use.
- Notes:
- New fields can be added to the OOB model or to a newly configured model, however additional configurations need to be implemented in Recommended Actions to support the fields display in Service Operations Workspace
- Totally new models can also be configured from the Admin Console which will need to be configured in Recommended Actions as well so that the predictions will display in Service Operations Workspace
- Notes:
- Observe the Input fields which are set to Short Description and Description in the OOB shipped model. These fields determine the input values from the incident where the model is predicting the output fields as prediction outcomes.
- Observe the Number of records count which are the total number of records that will be trained for the model once training is launched.
- Once the configurations are finalized, click on the Launch training button. Depending on the number of records being trained, it can take up to 30 minutes to a little over an hour for training to complete.
- Note: On production instances, the OOB model is auto-trained upon plugin installation when minimum data requirement is met. As a result, upon plugin activation, once the model is deployed in monitoring mode, the system will start predicting outcomes in monitoring mode in the background.
- Once training completes, the Assess your model page is displayed on Admin Console. The system automatically provides a field level assessment of Same as agent considered as correct predictions, Different which are considered as incorrect, and Skipped predictions. We can review each output field in more detail by clicking on the View sample results button.
Note: In case certain fields are not performing as expected, the model can be re-trained from Train your model page after making filter condition adjustments here. Refer to Community article Testing “Field Predictions” in Task Intelligence for tips on how to go about testing and improving model performance in Task Intelligence.
- Once the model is finalized, proceed to Set your field preferences tab as shown to either predict in the Service Operations Workspace agent user interface (as recommendations or auto-filled predictions) or only predict in the background in monitoring mode.
- Next, Deploy the model by clicking on the Deploy button from the Deploy your model page.
Configure Recommended Actions (RA):
TIP: Recommended Actions come pre-configured for the 3 output fields that are part of the OOB shipped model. Below pointers are meant to be a checklist to confirm RA is fully setup and provide guidance incase new fields need to be added to the OOB shipped model or a new model needs to be deployed. Refer to Community article Configure Additional Fields and Models in Recommended Actions for Task Intelligence ITSM for detailed guidance around this.
- As soon as your model is deployed, you will get the ‘Your model is deployed’ confirmation popup. Click on Configure Recommended Actions to go to Recommended Actions. This takes you to Recommended Actions>Context module on instance.
- Open Incident context Recommended Action Context record which comes pre-configured
- In the context record, under the Rules related list, ensure Active incidents Rule is set to Active=True
- In the context record, under the Recommendations related list, ensure Incident Fields value prediction (TI) recommendation shipped OOB is active. This recommendation is associated to OOB shipped Resource Generator called Incident Fields predictions TI that is linked to OOB shipped TI model ITSM OOB TI Solution configured for the 3 OOB model output fields: Assignment Group, Configuration Item, and Service fields.
- Any new output fields added to the OOB shipped model need to have their own Field Recommendation Actions, Field Recommendation Inputs, and Field Recommendation Definition settings configured. Detailed steps are provided in Community article Configure Additional Fields and Models in Recommended Actions for Task Intelligence ITSM.
- If a new/different-than-OOB model needs to be trained and deployed, its best to leave the OOB RA configurations as is and create a new Recommendation, linked to a new Resource Generator record (that is linked to the new model) with its own Field Recommendation Actions, Field Recommendation Inputs, and Field Recommendation Definition settings. Detailed steps around this are provided in Community article Configure Additional Fields and Models in Recommended Actions for Task Intelligence ITSM.
Predict incident field values to reduce handle time
- Use your data to predict the value of fields on new incidents.
- The prediction is made when an incident is created or updated.
- Prediction results are captured in ml_predictor_results_task table
- It can be deployed as recommendation, autofill or in the background only.
- The model supports 16 languages: English, Arabic, Chinese-simplified, Chinese-traditional, French, German, Italian, Japanese, Korean, Dutch, Polish, Portuguese, Spanish, Thai, Turkish, Russian
- Minimum data (as set in system property: glide.platform_ml.api.csv_min_line) needed for training is 10K records. Refer to Training Task Intelligence Models with Small Data Sets for guidance on how to update this minimum data requirement.
- Maximum allowed data (as set in system property: glide.platform_ml.api.csv_max_line) is 300K records.
- Output fields to be predicted must have at least 2 classes (distinct values) and at least 35 records in train data for each class
Supported Input Field Types:
Task Intelligence for ITSM supports the following field types (as allowed Inputs to the model during predictions) that can be selected as an Input field in the Admin Console’s Training tab (that is accessed from Task Intelligence Setup module) as shown in the image here.
INPUT: ['string', 'string_full_utf8']
Supported Output Field Types:
Task Intelligence for ITSM supports the following field types (as allowed Output fields predicted by the model) that can be selected as an Output fields in the Admin Console’s Training tab as shown in the image here.
OUTPUT: ['reference', 'integer', 'int', 'float', 'long', 'longint', 'choice', 'string', 'string_full_utf8']
|
Documentation
|
|
|
Additional Resources:
|
Check out the AI & Intelligence forum to get the latest updates on ServiceNow's AI products!
Hi @nilimadesai We have enabled Task Intelligence for ITSM and deployed the OOB model. Only admins are able to see predicted field values in Service Operations Workspace but agents with itil role are not able to see them. Should users have any additional roles other than itil to see predicted field values?
Task Intelligence gives errors as we are trying to develop/train a model in my Personel developer instance, we don't know the reasons. Please let me know how I can resolve this issue
