The Triage Cases Agentic Workflow in Now Assist for CSM automates the first step of every service interaction by handling customer inquiries intelligently and consistently. It coordinates multiple AI Agents that gather missing data, validate documents, respond to informational or transactional requests, and escalate complex cases to the right type.
This agentic workflow reduces manual effort by using contextual grounding, Knowledge Graph validation, and role-specific orchestration. It dynamically adapts to each situation while maintaining deterministic structure for accuracy, governance, and auditability.
Key Capabilities
-
Context understanding: Uses sentiment and intent to determine the best triage path.
-
Entity extraction: Captures required data from emails or attachments automatically.
-
Document verification: Validates case documentation based on configured business rules.
-
Case creation and routing: Creates or links cases intelligently to prevent duplicates.
-
Automated responses: Handles informational or transactional questions using grounded knowledge.
-
Wrap-up summarization: Compiles triage actions for reporting and follow-up consistency.
Overview video:
List of AI Agents included as part of Triage cases (as of FY25 Q4):
Implementation
1. Triggering the Workflow
When an inbound email or case is created, the Triage Cases Agentic Workflow begins automatically. Note that the trigger point is configurable. A notification in the Now Assist Panel informs the agent that triage is in progress.
Best practices on email reply triggers:
- Avoid Creating AIA Triggers on
sys_emailTable: The AI Agents team has deprecated support for creating AIA triggers directly on thesys_emailtable. - Use Flow Designer with Inbound Email Triggers: Instead of relying on multiple inbound email triggers (which are limited by design), integrate the AIA use case into existing flows.
- Manual Flow Action Addition:
- Open Flow Designer.
- Access the flows:
"update interaction from email"and"UpdateCase via Reply for EaaI". - At the final step of each flow, click the plus icon to add a new action.
- Add the action:
"Trigger AIA use Case from email". - Set the email variable to:
Trigger inbound email > email record. - Publish the updated flows to activate the changes.
1.1 Validate the data
The Context Validator AI Agent reviews the record:
-
Validates the user identity and record type
-
Runs early sentiment and intent analysis for short-term memory use
-
If the record is invalid or missing required information, the process pauses and notes are added so a human agent can take over
1.2 Sentiment analysis
Generative AI evaluates the customer’s tone:
-
If the sentiment is negative, the case is escalated for human handling
-
If positive or neutral, the workflow continues automatically
1.3 Intent analysis
The system determines what the customer is trying to achieve (for example, asking for information or requesting an action).
2. Informational intent
If the customer is looking for information or has a specific question:
-
The Informational Queries AI Agent searches for contextual answers using Knowledge and Knowledge Graph sources
-
The Email Response AI Agent sends the generated reply back to the customer
3. Transactional intent
If the request requires an action, such as a refund or status update:
-
The Entity Extraction AI Agent identifies key details from the message (for example, product name, issue type, or ID) and maps them to the case fields
-
Optionally, the Document Verification AI Agent validates attached documents
-
The Email Response AI Agent sends the confirmation or next-step instructions
As this is a more advanced AI Agent, it requires more tool configuration for its customization. To customize entity extraction for Case types:
1. Go to the AI Agent Studio, open the Entity Extraction AI Agent and then open the "Get whitelisted fields and task definition" tool (under Scripts) you will find the NowAssistCSMAlAgentConstantsSNC script include.
2. Open it and find or add the EXTRACTION_TASK_DEFINITION object like this:
3. Then define the case type with sys ID like this (see Complaint Case as an example):
4. Define below the white list fields map per your use case in this format (do not use natural language here):
4. Case creation
If the issue is too complex for automated resolution:
-
The Case Creation AI Agent opens a new case record aligned to the organization’s configured case types
-
It checks for related knowledge articles or existing cases
-
If it’s a repeat issue, the new case is linked to the existing one
-
If not, the case progresses through the next validation steps
-
5. Verify documents
The Document Verification AI Agent inspects all attached files or forms to ensure completeness and compliance.
-
If mapping from the document or file is required, the workflow returns to the Entity Extraction AI Agent for further mapping into case fields.
As this is a more advanced AI Agent, it requires more tool configuration for its customization. To customize document verification for Case types:
- Go to Checklist Templates table (example is provided with OOTB Onboarding case type)
3. Open the NowAssistCSMAlAgentConstantsSNC script include. Modify as needed with the specific template sys ID or add your own Case types here (see lines 10-12 as an example)
Note: If curious how is it used, just go to AI Agent Studio, within the Documents Checklist tool in the AI Agent you will find "CSMAgentDocumentUtils" script include that does a call through getTemplateId to it.
6. Send a response
The Email Response AI Agent drafts and sends a message to the customer confirming the completed actions or outlining what will happen next.
7. Close the loop
The workflow completes by updating the case record with all validated details, outcomes, and audit logs.
The WrapUp AI Agent summarizes every action taken and adds it to the case activity stream. If additional support is needed, the case is seamlessly handed off to a live agent with full context preserved.
-
Note that you don't need to use all parts of the agentic workflow at the beginning.
Admin Experience walkthrough:
Live Agent Experience walkthrough:
Key Best Practices
-
Start with documenting your current process in a diagramming tool and iterate from there. Usually triage processes vary by service or product offering.
-
Remember our agentic workflows are simply models or examples for you to ground/clone using AI Agent Studio
-
Customize directives for sentiment or verification logic instead of modifying the entire orchestration.
-
Use Knowledge Graph grounding to reduce response variability and maintain compliance. Connect external platforms as needed.
-
Monitor metrics in AI Agent Studio Analytics to identify which agents perform best.
-
Schedule regular refinement sessions to adapt intent and document logic as new case types evolve.
Formal Learning
Measured Success
| Outcome | Explanation (with applicable use case) | Success Metric |
|---|---|---|
|
Reduced manual triage workload |
Automates validation, classification, and case creation from inbound requests |
% of cases auto-triaged |
|
Increased routing accuracy |
Uses sentiment and contextual reasoning to determine correct case type |
% of correctly routed cases |
|
Shorter resolution cycle |
Enables early classification and response to simple transactional requests |
Average time to triage (minutes) |
|
Fewer duplicates |
Links similar inquiries and prevents unnecessary case creation |
Duplicate reduction (%) |
|
Improved data quality |
Extracts consistent customer and case data across all channels |
Field completion rate (%) |
Frequently Asked Questions (FAQs)
1. How do I clone and customize the triage workflow?
Admins can clone the workflow in AI Agent Studio using the “Clone AI Agent Use Case” option, then edit or replace AI Agents to tailor the orchestration.
2. Can I reuse triage agents in other workflows?
Yes. Each AI Agent (such as Context Validator or Entity Extraction) can be reused or extended across multiple workflows.
3. How can I test changes safely?
Clone the workflow into a non-production environment, validate data quality, and use sample emails or cases before activation.
4. What data sources are required for grounding?
Beyond good prompting, Knowledge Graph, transactional records, and complete foundational data models are required for contextual grounding and data accuracy.
5. How is orchestration maintained?
Each AI Agent follows deterministic sequencing defined in AI Agent Studio through the agentic workflow orchestrator. Admins can reorder steps or assign triggers without coding and even use generative AI to create their own instructions, categories, get suggestions of tools and AI Agents.
6. What metrics should I monitor first?
Focus on auto-triage rate, duplicate case reduction, and mean time to first response, Average Handle Time (or First contact resolution) for the most direct indicators of efficiency.
7. Why is my Informational Queries AI Agent not finding knowledge articles?
The title from Knowledge Article drives the major result through the Get relevant knowledge articles RAG tool. If you want to be more flexible in the match, go to Document matching threshold and change it from 0.8 (strict match) to a lower one (from 0 to 1). Higher is more strict in the results matching.
