FernandoCastro
ServiceNow Employee
ServiceNow Employee

Agentic workflows extend Now Assist for CSM beyond human assistance by enabling autonomous execution of multi-step actions directly from the context of a case, conversation, and/or customer intent, independent of where the data resides thanks to AI Agent Fabric and Workflow Data Fabric.

These flows reduce friction for agents and allow the platform to act on their behalf using secure guardrails.

Prebuilt agentic workflows like Triage Cases or AI Agents like Troubleshooting Steps identification leverage the full ServiceNow AI Platform (Flow Designer, Script, Topics, Catalog Items, RAG, Record operations, web search, Generative AI inputs and more) and any existing automation you've build to intelligently guide or execute behind-the-scenes processes while mix and matching solutions for your specific customer facing teams. The combinations are up to your use cases. This allows humans focus on more complex work to increase customer satisfaction and resolution.

 

Included Capabilities:

  • Triage Cases: Dynamically responds to interactions or cases based on customer sentiment and intent while gathering context, verifying user data, responding informational or transactional queries or escalating to complex Cases types if needed 

  • Overview video:

 

List of AI Agents included as part of Triage cases:

Screenshot 2025-08-04 at 7.56.12 PM.png

 

Overall flow:

tcoverallflow.png

1. New case arrives

A customer message comes in (via email, chat, or form). This starts the automated workflow (this is configurable).

2. Plan the triage path

The Triage Cases Planner AI Agent determines which AI agents and steps should run based on the context of the use case

3. Validate the data

The Context Validator and Analyzer AI Agent reviews the record:

  • Checks if the data is complete

  • Verifies the user 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 stops and notes are added so a human agent can takeover.

4. Identify duplicates

The Duplicate Record Identifier AI Agent checks whether this is a repeat issue:

  • If yes: link it to the existing case

  • If no: continue with the triage process

5. Sentiment analysis

AI evaluates the customer’s tone:

  • If negative, the case is escalated and handled by a human

  • If positive or neutral, the process continues

6. Intent analysis

AI determines what the customer is trying to do:

  • If they just want information → handled by the Informational Queries AI Agent, and then sent via Email Response AI Agent

  • If it’s an action (like a refund or update) → handled by the Transactional Queries AI Agent, optionally have some Documents verified and finally send a response via Email Response AI Agent

7. Case creation

If the request is too complex for auto-resolution, the Case Creation AI Agent opens a full case record in the system following specific configured Case Types that the instance has available.

8. Extract key details

The Entity Extraction AI Agent pulls out important facts (attributes like product names, issue types, or IDs) from the customer message and maps them into the case type fields.

9. Verify documents

The Document Verification AI Agent checks any uploaded files or attachments to make sure they’re complete and valid. Some of these can also go back to Entity Extraction AI Agent to map them into case type fields.

10. Send a response

The Email Response AI Agent drafts and sends a reply to the customer, confirming what was done or what to expect next.

11. Close the loop

The system updates the case with all results and either resolves it or passes it to a human agent for further action.

 

Admin Experience walkthrough:

 

Live Agent Experience walkthrough:

  • Troubleshooting Steps Identification (AI Agent - not a workflow): Gathers context from cases, identifies missed fixes by comparing with knowledge bases, similar cases, and standard operating documents, and proposes additional troubleshooting steps.

Key Best Practices

7 Proven Practices for Successful Agentic AI Implementation

  1. Design with Structure First
    Give each step a single owner (agent) and a single tool. Streamline parallel processes into one clear flow and trigger them in predictable ways to reduce complexity. Reasoning prompting excels at this.

  2. Prioritize the User Experience
    Make interactions seamless by pulling data from existing records, requesting inputs in intuitive ways, and keeping each step focused on one clear tool while allowing AI agents to dynamically choose the right tool at runtime.

  3. Create Standards You Can Trust
    Use consistent naming, formats, and instructions so AI agent behavior is predictable and traceable. Enforce data privacy rigorously, especially when handling sensitive information.

  4. Don’t Wait for Perfect Data
    Launch with out-of-the-box, productivity-focused agents to deliver quick wins and using generative AI to clean and structure key data (Knowledge, resolution notes, CRM Foundation data models, others). Early successes feed better data back into the system, accelerating future improvements.

  5. Check Readiness Before You Deploy
    Use instance readiness tools to detect customization conflicts before rollout. Preventing issues is faster and easier than fixing them post-deployment.

  6. Know When Agentic AI Isn’t the Answer
    Not every task needs full Agentic AI. Reserve it for problems that require reasoning and planning. Keep single-step or simpler tasks in Now Assist skills or Flow Designer to avoid unnecessary complexity.

 

Measured Success

How do we usually measure success in this set of purpose-driven skills?

Outcomes Explanation (with applicable use case) Success Metric

Faster time-to-triage

Reduce time spent manually classifying or reassigning cases by auto-executing routing flows Use case: Triage Cases

Avg. time from case creation to routing (seconds)

Reduced agent intervention

Decrease manual effort by letting agentic flows execute background tasks when conditions are met Use case: Troubleshooting Steps

% of tasks executed autonomously

Higher accuracy in resolution

Deliver more consistent troubleshooting flows by embedding decision logic at point of need Use case: Troubleshooting Steps

% of cases resolved without additional escalation

Increased workflow reuse

Use modular agentic flows that can be reused across departments or processes with case types Use case: Triage Cases

Number of flows reused across workflows or LOBs

Reduced mean time to resolve

Accelerate case resolution by acting on contextual triggers rather than waiting for agent input Use case: Triage Cases + Troubleshooting Steps

MTTR (minutes)

 

Version history
Last update:
4 weeks ago
Updated by:
Contributors