What is Virtual Agent?
Virtual Agent (VA) is ServiceNow’s conversational chatbot framework. Customers ask questions in natural language, and the bot answers, gathers information, creates or updates cases, or hands the conversation to a live agent. It uses Natural Language Understanding (NLU) to detect what the user means.
| Component | Description |
| Topic | A conversation flow built in Virtual Agent Designer (for example, Check case status). |
| Topic block | A reusable piece of conversation embedded inside topics. |
| Intent | What the user wants, defined in an NLU model. |
| Utterance | A sample phrase a user might type for an intent. |
| Entity | Data extracted from text, such as a case number or product. |
| Channel | Where the bot appears: portal, Workspace, Teams, Slack, SMS, and others. |
| Live agent handoff | Transfer to a human through Agent Chat and Advanced Work Assignment. |
Why Use Virtual Agent in CSM?
- Deflection: resolve common questions without an agent.
- 24/7 availability and immediate answers with no queue.
- Lower cost to serve.
- Better data capture: structured details are collected before an agent joins.
- Consistent answers from approved knowledge.
When to Use Virtual Agent?
Good fit
- High-volume, repetitive requests such as case status, order tracking, and FAQs
- Simple case creation with structured inputs
- After-hours support
- Triage before handoff to an agent
Poor fit
- Complex, emotional, or high-value issues that need a human
- Situations with no quality knowledge or stable process to automate
Introduce Virtual Agent after the case model, knowledge base, and portal are stable. A bot is only as good as the content and processes behind it.
How to Implement Virtual Agent
Step 1: Prerequisites
- Activate Virtual Agent, NLU Workbench, Agent Chat, and Advanced Work Assignment as needed.
- Assign virtual_agent_admin to designers.
- Ensure knowledge articles are published and searchable.
Step 2: Plan Use Cases
- Analyze case data to find the top contact reasons.
- Prioritize by volume, simplicity, and business value.
- Document the conversation flow and success criteria for each use case.

Step 3: Enable Out-of-the-Box CSM Topics
- Review the provided CSM topics (such as create a case, check case status, search knowledge) and activate the ones needed.
- Clone system topics before modifying them so upgrades do not overwrite changes.

Step 4: Build Topics in Virtual Agent Designer
Navigate to Conversational Interfaces > Virtual Agent > Designer, create a topic, and add nodes:
- Bot response: messages, images, and links.
- User input: text, choices, dates, and reference lookups.
- Utility: run scripts, call flows, look up records, and create or update cases.
- Decision: branching logic.
- Topic block / Handoff: reuse logic or transfer to a live agent.
Step 5: Train the NLU Model
- Open NLU Workbench and create a model.
- Define intents that match your topics and add many varied utterances for each.
- Add entities where needed.
- Train, test, and publish the model.
- Map intents to topics.

Step 6: Configure Channels
- Add the Virtual Agent widget to the Customer Service Portal.
- Configure additional channels (Teams, Slack, SMS) as required.
- Set the welcome message, branding, and the fallback message for unrecognized input.
Step 7: Configure Live Agent Handoff
- Set up AWA queues, agent skills, and presence.
- Pass the conversation transcript and collected details to the agent Workspace.
- Define business hours and offline behavior (for example, create a case if no agent is available).
Step 8: Test
- Test each topic, including wrong or unexpected inputs.
- Test NLU accuracy with phrases the model has not seen.
- Run user acceptance testing with real support staff.

Step 9: Deploy and Monitor
- Release in phases, starting with a limited audience.
- Review Virtual Agent analytics: conversation volume, containment, handoff rate, and failed intents.
- Continuously add utterances and topics based on unresolved conversations.
Sample Virtual Agent Use Cases
| Use case | Bot behavior |
| Check case status | Ask for the case number, look it up, and return the status. |
| Create a case | Collect product, issue, and description; create the case; confirm the number. |
| Knowledge search | Answer from knowledge and ask whether it helped. |
| Update a case | Add a comment or attachment to an existing case. |
| Escalate to agent | Summarize context and transfer to live chat. |
| Collect feedback | Present a CSAT survey at the end. |
Virtual Agent Best Practices and KPIs
- Start with 3 to 5 high-value topics and expand.
- Keep messages short and conversational.
- Always offer a path to a human.
- Handle failure gracefully with clear fallback messages.
- Reuse topic blocks to avoid duplication.
- Authenticate users before showing account-specific data.
- Regularly review failed conversations and retrain the model.
KPIs: containment / deflection rate, handoff rate, fallback (not understood) rate, average conversation length, resolution rate and CSAT, and top intents and top failed intents
