AI / ML integration

Wire AI/ML into workflows two ways: native Predictive Intelligence / Now Assist in-platform, and calls to an external model over REST for custom scoring.

• ServiceNow: admin; for native ML the Predictive Intelligence plugin; for GenAI the Now Assist subscription.
• For external models: an endpoint URL and API key, plus IntegrationHub for the REST call.

• Go to Predictive Intelligence → Solution Definitions and create a Classification solution.
• Pick the table (incident), the input field (short_description) and the output (category or assignment_group).
• Set training filters (e.g. last 12 months, resolved) and train the solution; review precision/coverage.
• Activate it and add a business rule / flow that predicts on insert to auto-categorise and route.

• Enable Now Assist skills (incident summarisation, resolution notes).
• Drop the Now Assist panel on the incident form; agents get one-click summaries of long records.

• Create a Connection & Credential alias for the model endpoint with the API key.
• In a flow, add a REST step that sends record fields and reads back a score/label.

POST https://api.example-ml.com/v1/score
Authorization: Bearer {api_key}
{
  "text": "{short_description}",
  "features": { "priority": "{priority}", "ci": "{cmdb_ci}" }
}
// response: { "risk_score": 0.82, "label": "high" }

• Write the returned risk_score back to the record and branch the flow (e.g. add approvers when high).

• Create test incidents and confirm predictions/scores populate and routing changes.
• Re-train native solutions on a schedule as data grows; monitor prediction quality.

• Low coverage: not enough clean training data, widen filters or improve labels.
• External 401/timeout: check the credential and endpoint, and consider async for slow models.