ServiceNow Otto for CSM - Live Agent Advisor

FernandoCastro · 3 hours ago

Give voice agents real-time, cited AI guidance without leaving the call

Live Agent Advisor (also called Live Agent Assist) is an AI Agent that reads the live transcript of a voice interaction, retrieves cited Knowledge Base answers through AI Search, and pulls live account and product context through the Enterprise Knowledge Graph. It then surfaces a single, cited recommendation in the Otto panel of the Agent Workspace, replacing manual tab-switching across knowledge and account systems mid-call with a one-click, human-in-the-loop retrieval.

Live Agent Advisor is delivered as an AI Agent through the Customer Service Management AI Agent Collection plugin. It requires a supported CCaaS integration (Genesys, Twilio, NICE, Five9, 3CLogic, or Amazon Connect) with real-time transcription enabled, and its trigger must be activated separately in AI Agent Studio; it is not on by default once the plugin is active.

Key Capabilities:

Automatic trigger on live interaction: Live Agent Advisor activates the moment a live human agent — not a virtual agent or bot — picks up an active phone call with a customer identified, with no manual setup per call.
One-click cited recommendation: Clicking Generate analyzes the live transcript, runs an AI Search against configured Knowledge Base articles, and returns a recommendation in the Otto panel with every claim tied to a citable source.
Live account and product context: The Enterprise Knowledge Graph pulls account, contract, asset, entitlement, and install-base data for the current caller and combines it with the KB answer, rather than surfacing a generic article alone.
Update as the call evolves: Clicking Update refreshes the recommendation with the latest transcript context as the conversation continues.
Natural-language follow-up: Agents can type a follow-up question (for example, "what plan is this customer on?") and the AI Agent re-runs its retrieval with that context while maintaining conversational continuity across the call.
Multi-query and clarification handling: If a customer asks more than one question at once, the agent is prompted to choose which to address first. If the AI Agent needs more context before answering, it presents clarifying questions as selectable options in the Otto panel and waits for the agent's choice.
Human-in-the-loop throughout: Agents can stop AI processing at any point; the AI Agent never acts on the customer's behalf, it only surfaces a recommendation for the agent to use.
Nothing to close out per call: Live Agent Advisor's session ends on its own when the call itself closes — there's no separate step for the agent or an admin to shut it down each time.

See It In Action

Telecommunications: A customer says their internet's been down since this morning and asks in the same breath whether they're still under contract. Live Agent Advisor surfaces both open questions as quick options for the agent to pick from, then returns cited troubleshooting steps grounded in the customer's actual broadband plan — no separate lookup, no hold.

Healthcare & Life Sciences: A member asks if an upcoming procedure is covered and what they'll owe. Live Agent Advisor combines the relevant coverage policy with the member's own plan details into one answer — not a policy citation the agent still has to match to the account by hand.

Retail: A customer reports receiving the wrong item and asks about the return window before deciding what to do. After Live Agent Advisor answers the return-window question, the customer asks a quick follow-up — could they exchange it instead? — and the AI Agent keeps the answer anchored to that same order without the agent re-explaining the situation.

Implementation Steps

From All > AI Agent Studio > Create and Manage, select Live Agent Assist in the AI Agents tab (the in-product label at this writing; see the Naming FAQ below). Open the Add triggers section, select the Live Agent Assist trigger (inactive by default), turn on "Trigger is ON," and save. Configuring or activating the trigger requires the now_assist_admin role. Full setup and data-source detail is in the documentation below. The out-of-the-box configuration is a starting example, not a finished thing to activate as-is: clone the AI Agent, then adapt its data retrieval (the AI Search Profile and Knowledge Graph tag), its action tools, its context scoping, and its prompt/instructions to your own Knowledge Base, account model, and policy content, and run evals against that grounded configuration before turning the trigger on for live agents handling real customer calls.

Best Practices

Confirm real-time transcription before activating the trigger: Verify your CCaaS integration has real-time transcription enabled first; the AI Agent has no transcript to read without it.
Start with the default AI Search Profile scope: The AI Search Profile is scoped to the Knowledge (kb_knowledge) table by default; confirm your relevant KB articles live there before assuming coverage gaps.
Extend context through the Knowledge Graph tag, not a custom build: Add tables to the Enterprise Knowledge Graph tag (Knowledge Graph Designer > Enterprise Graph > CSM Live Agent Assist Tag) for additional context first; reserve a custom Customer Context Builder implementation for fields the tag approach genuinely can't reach.
Confirm data privacy plugin configuration before rollout: Masking applies to data sent to the LLM during processing only if the data privacy plugin is correctly configured on your instance. Verify this before go-live, not after.
Train agents on the Generate / Update / Stop workflow: This changes the live agent's in-call workflow; walk agents through the Otto panel controls and the multi-query clarification flow before broad rollout, not just at go-live.

Formal Learning

ServiceNow University: Now Assist for CSM Implementer Learning Path

Measured Success & Outcomes

Outcome	How It Supports Teams	Key Metric
Lower average handle time	Removes 30–90 seconds of manual searching across knowledge and account systems per query	Average handle time (AHT) on voice calls
Improved first-contact resolution	Gives newer agents the same instant, cited guidance tenured agents rely on instinct for	First-contact resolution rate
Fewer repeat contacts	Grounds answers in the caller's actual account and product history instead of generic KB content	Repeat-contact / callback rate

FAQs

Q. Is this "Live Agent Assist for Voice" or "Live Agent Advisor"?
Live Agent Advisor is the approved name for this capability going forward. Some in-product labels, including the AI Agent Studio trigger name and the Knowledge Graph tag name, still say "Live Agent Assist" at this writing; that reflects a not-yet-completed rename of those configuration artifacts, not a different feature.

Q. What do I need in place before I can turn this on?
The Customer Service Management AI Agent Collection plugin must be active, you need a supported CCaaS integration (Genesys, Twilio, NICE, Five9, 3CLogic, or Amazon Connect) with real-time transcription enabled, and the Live Agent Assist trigger must be activated in AI Agent Studio. Activation requires the now_assist_admin role.

Q. Which roles can use it?
sn_customerservice_agent for B2B agents and sn_customerservice.consumer_agent for B2C agents.

Q. Where does it pull its information from, by default?
Three sources: the Customer Context Builder (a direct database query for high-priority fields like recent interactions and installed products), the AI Search Profile (scoped to the Knowledge kb_knowledge table for KB article retrieval), and the Enterprise Knowledge Graph (covering account, asset, case, consumer, contact, contract, entitlement, install base item, installed product, interaction, product model, and sold product data).

Q. Can we add more data sources?
Yes. Adding tables to the Enterprise Knowledge Graph tag (Knowledge Graph Designer > Enterprise Graph > CSM Live Agent Assist Tag) is the supported, no-code path. Extending the Customer Context Builder itself requires a custom scripted implementation of its interface class, and should be reserved for fields the tag approach can't reach.

Q. Does it stream its response, or remember earlier calls?
No. Recommendations render complete rather than token-by-token, and the AI Agent is stateless by design: each invocation reads only the current call's live transcript, with no memory carried across calls beyond that.

Q. Is customer data masked before it reaches the model?
Data sent to the model during processing can be masked, provided the data privacy plugin is correctly configured on your instance. Once an agent copies AI-generated content into a case's work notes, that copy is no longer masked. From that point it's a standard, unredacted instance record governed the same way any other work note is.

Q. What license tier is this under?
CSM Advanced.

Ready to go deeper?

Head back to the Implementation Library Home for the full guided path across Now Assist for CSM.

Back to Implementation Library Home