- Post History
- Subscribe to RSS Feed
- Mark as New
- Mark as Read
- Bookmark
- Subscribe
- Printer Friendly Page
- Report Inappropriate Content
5 hours ago
ServiceNow AI Voice Agents
GenAI-powered voice agents built for natural, dynamic phone conversations that resolve issues in real time
What Are AI Voice Agents?
AI Voice Agents are intelligent conversational systems that understand, respond, and act through natural spoken interactions. Unlike traditional IVR systems that force callers through rigid touch-tone menus, AI Voice Agents use large language models (LLMs), real-time speech recognition, and voice synthesis to have natural conversations — and take action on the caller’s behalf.
They deliver support across voice-enabled channels including phone (via CCaaS), web (WebRTC), and mobile — handling everything from checking case status to managing order operations to creating new cases, all without requiring a customer service rep.
Key Distinction: AI Voice Agents resolve issues — they don’t just route calls. They understand intent, access customer data, execute workflows, and deliver outcomes through natural conversation.
Why Voice Still Matters
Despite a decade of digital transformation, voice remains the dominant support channel. Customers continue to reach for the phone — especially when issues are urgent, complex, or emotionally charged.
of customers choose voice even when digital options exist
channel for urgent, complex, and high-emotion interactions
of customers say phone is the fastest path to resolution
The Cost Problem: Voice interactions cost $8-12 each compared to $1-3 for chat, and 70% of contact center spend is tied to voice operations. AI Voice Agents close this gap — resolving calls at self-service cost with customer service rep quality.
The Evolution of Voice in Customer Service
Voice technology has progressed through three distinct generations — from rigid touch-tone menus to intelligent, action-taking agents:
“Press 1 for billing”
Touch-tone keypad input only. Hardcoded decision trees with zero personalization. Required separate IVR per language. Could only route to queues — humans resolved everything. 80% of callers hung up in IVR.
“Say billing or support”
Spoken word input with limited vocabulary (Google Dialogflow, Amazon Lex, Nuance). Intent classification and slot filling. Could handle simple FAQs and routing but still required rigid flows and per-model language config.
“How can I help you today?”
Natural conversation with any phrasing. LLM reasoning with context and memory. Knows cases, orders, and history. Multilingual by default. Resolves issues by creating, updating, and acting on records. Maintained with natural language instructions.
Two Sides: Self-Service & Agent Assist
AI Voice Agents operate on both sides of the customer interaction — resolving issues autonomously for callers and empowering customer service reps when escalation is needed:
Self-Service (Customer-Facing)
The AI Voice Agent handles the caller’s request end-to-end without involving a customer service rep:
- Containment — Resolve issues completely within the voice interaction
- Workflow Automation — Execute platform actions (create cases, update records, look up orders)
Agent Assist (Agent-Facing)
When calls transfer to customer service reps, Now Assist enhances their effectiveness:
- Coaching — Real-time guidance and suggested responses
- Recommended Actions — AI-suggested next steps based on conversation context
Core Capabilities
Powered by real-time speech recognition (Whisper Large 3), voice synthesis (Cartesia Sonic 3), and LLM intelligence. Supports English, German, Spanish, and French with additional Tier 1 & 2 languages coming.
Seamless integration with Genesys Cloud and Twilio at launch, with 3CLogic, Amazon Connect, Five9, and NICE support planned for future releases.
When AI can’t resolve an issue, calls transfer to customer service reps with full conversational context — or automated records are created to ensure nothing is lost.
Personalize your agent with an extensive voice library. Configure persona, language, welcome message, and tone to match your brand.
Supports Azure OpenAI, Google Gemini, and Anthropic Claude on AWS — accessed via ServiceNow-managed private connections for security and performance.
Robust performance tracking through an enhanced analytics dashboard with actionable insights including deflection rates, CSAT scores, and full call transcripts.
Out-of-the-Box Use Cases
ServiceNow ships ready-to-use AI Voice Agents for common scenarios across ITSM, CSM, and HRSD. You can also build custom agents using natural language instructions in AI Agent Studio.
ITSM
- Create New Incident
- Check Incident Status
- Outlook Troubleshooting
- Catalog Item Request
- Password Reset
HR
- Create New Case / Check Case Status
- PTO Related Inquiries
- Travel Expenses
- Update Employee Information
CSM
- Manage Order Operation
- Order Case Self-Service
- SOM — Order Exception Management
- Invoice Exception AI Agent
- RMA AI Agent (Returns)
- CPQ: Quote AI Agent (includes voice chat)
- Build Your Own via AI Agent Studio
How It Works: Architecture & Data Flow
AI Voice Agents run within the ServiceNow Commercial Cloud boundary. The architecture consists of three main zones: your CCaaS provider handling telephony, the ServiceNow Compute Hub running AI voice services, and your customer instance providing data, workflows, and tools.
End-to-End Call Flow
Every voice interaction follows this sequence from spoken word to resolved outcome:
Key Components
Hosted in the ServiceNow Compute Hub. Manages the full call lifecycle including routing, turn-taking, agent selection, authentication, context gathering, and tool orchestration.
Azure OpenAI, Anthropic Claude on AWS, and Google Gemini for reasoning. Accessed via ServiceNow-managed private connections to provider-hosted infrastructure.
Flows, subflows, scripts, search retrieval, knowledge graphs, MCP server tools, and record operations — all executing within the customer instance with ACL enforcement.
Short-term memory tracks current task and conversation history. Long-term context includes customer data, case history, and knowledge base content accessed via RAG.
Security: All flows use TLS 1.2 encryption in transit. The customer instance supports domain separation. Tools execute with ACL-based access control.
Voice Agent Configuration
Configuration uses two tools working together: Assistant Designer for the overall voice experience, and AI Agent Studio for individual agent behaviors.
A. Voice Personality
Language, welcome message, and persona (e.g., Receptionist, Southern Woman)
B. Authentication
Caller ID (phone/email), auth type (multi-factor or single), method (TOTP, SoftPIN, SMS OTP, Push)
C. Safeguards
Fallback behavior (customer service rep or create ticket), max call duration, inactivity timeout, detail capture toggle
D. CCaaS Provider
Select Twilio (Webhook) or Genesys (Audio Connector). A URL is generated for your CCaaS config.
Option A: Use Out-of-Box Agents
ITSM (password reset, ticket status), HR (employee self-service), CSM (customer service workflows)
Option B: Build Your Own
Write natural language instructions, define specialty and steps, set tool access and user access rules, configure entity context
Tools & Capabilities
Map tools: knowledge search, record lookup, approvals, flows, scripts, RAG search, MCP, file upload. Each Voice Agent handles one topic — add multiple per assistant.
Final Step: Map all applicable Voice Agents to the Voice Assistant, click Activate to deploy, then test via a live call. The activation generates a URL to configure in your CCaaS provider.
Tools Available to AI Voice Agents
AI Voice Agents leverage existing platform tools plus new additions. Here’s what’s available and when to use each:
| Tools | When to Use |
|---|---|
| Flow Actions, Subflows | Execute logic, update records, launch processes. For repeatable actions or structured logic. |
| Now Assist, Custom Skills | Generate, summarize, or classify text. For AI-powered content generation within workflows. |
| Script, Record Operation, Catalog Item, Conversational Topic | Handle platform logic or invoke external systems. When actions require custom logic. |
| Search Retrieval, Knowledge Graph, File Upload | Provide background context for LLM decision-making. When agents need context to reason (e.g., related incidents). |
| Web Search, MCP*, Agent-to-Agent (A2A)* / AI Agent Fabric* | Extend to third-party data and APIs. When context lives outside of ServiceNow. (*Coming soon) |
Security, Guardrails & Compliance
AI Voice Agents include multiple layers of security and governance built into the platform:
- Phone number identification
- Soft PIN (6-digit enrolled)
- TOTP (Authenticator app)
- SMS OTP (via Twilio)
- Push Notification (Okta Verify)
- Multi-factor support
- Now Assist Guardian built-in
- 16 categories (toxicity, bias, bad financial practices, etc.)
- Prompt injection protection
- Jailbreak detection
- Log-only or block-and-log modes
- PII masking before LLM calls
- Regex-based detection (email, phone, SSN, etc.)
- Placeholder sent to LLM, original text restored for user
- TLS 1.2 encryption in transit
- Role-based ACLs per agent
- Least-privilege execution
- Agent Evals (golden truth data)
- Run 1,000 simulations before production
- Analytics dashboard for monitoring
Mobile Deployment
AI Voice Agents can be launched from the ServiceNow Mobile App in two ways: directly via a dedicated voice button, or from within the Now Assist Virtual Agent conversation. Mobile channel support begins in Q1 2026.
Users tap a Prominent Action Button on the mobile home screen to immediately start a voice conversation.
Users can switch to voice mid-conversation from the Now Assist chat interface using a voice launcher button.
Prominent Action Button Availability: Australia Patch 0 (AP0), Zurich Patch 4 (ZP4), Yokohama Patch 11 (YP11). Only one Prominent Action Button is allowed per mobile client.
Key Details & Considerations
50 Assists per call. Calls shorter than 30 seconds don’t count toward consumption.
Genesys Cloud and Twilio supported at launch. Additional providers coming in Q1 2026. Omni-channel starting with Mobile App in Q1 2026.
Uses a separate Voice Orchestrator. Tool set within Voice Agents is slightly more limited than standard AI Agents. Includes flows, scripts, RAG search, MCP, and file upload.
English, German, French, and Spanish supported. Configured at the Voice Assistant level. Additional Tier 1 and Tier 2 languages coming soon.
Not available in Self-Hosted, In-Country, or NSC environments. No outbound calling. Audio input only (no multimodal). No real-time language detection. Manual testing (automated testing coming later).
How to Pick Your First Use Case
Selecting the right starting point is critical for a successful AI Voice Agent deployment. Here’s a practical framework:
✓ Look For
- High call volume (100+ calls/day on this topic)
- Repetitive, well-documented process
- Clear resolution path with defined outcomes
- Data already lives in your platform (CRM, ITSM)
- Current handle time > 5 minutes per call
✗ Avoid Starting With
- Highly emotional or complaint-heavy calls
- Processes requiring legal/compliance judgment
- Workflows that change frequently
- Topics with no structured data behind them
- Multi-party or complex handoff scenarios
Best Practices
✓ Do
- Start with one well-defined workflow
- Iterate with real production data
- Hold weekly demos to prevent scope drift
- Document decision points of each AI agent
- Plan for upskilling from day one
- Measure what matters to the business, not just the tech
- Write prompts an 11-year-old could follow
- Give the LLM a graceful “I don’t know” path
✗ Don’t
- Skip discovery to start building
- Over-engineer with too many agents
- Ignore negative scenarios and edge cases
- Expect 100% automation or identical results every time
- Use voice agents where a simple flow or IVR suffices
- Deploy without running Agent Evals first
- Forget human-in-the-loop for regulated industries
- Skip governance over workflow and prompt changes
Implementation Guide
The full implementation guide covers the end-to-end process for deploying AI Voice Agents. Here’s a summary of the key steps involved:
Ensure you have Now Assist licensing, a supported CCaaS provider (Genesys Cloud or Twilio), and the AI Voice Agents plugin activated on your instance.
Use Assistant Designer to set up voice personality, authentication method, safeguard behaviors, and CCaaS provider connection.
Use OOTB agents for ITSM, HR, or CSM, or build custom agents in AI Agent Studio with natural language instructions, tools, and access rules.
Assign Voice Agents to the Voice Assistant. Map required tools (flows, knowledge search, record operations) to each agent.
Run simulations using golden truth data (up to 1,000 scenarios) before production. Review analytics dashboard for performance tuning.
Click Activate to generate the CCaaS webhook/connector URL. Configure this URL in your Twilio or Genesys Cloud environment. Test via live call.
Full Implementation Guide
For detailed step-by-step instructions including CCaaS provider-specific configuration, mobile deployment setup, and advanced customization options:
View Implementation Guide →Frequently Asked Questions
Now Assist Voice Agents are architected for real-time workflow execution on the Now Platform. They minimize unnecessary system hops and have native awareness of CMDB, records, and workflows — meaning they can take meaningful action without heavy integrations. The result is faster, more natural, more contextual voice automation without the hidden cost of maintaining separate CCaaS voice bots.
You need a Now Assist license (Pro+ or Enterprise+). Required patch level is Zurich Patch 4 (also available in Yokohama Patch 11). Required plugins include nowassist-ai-agents, nowassist-aia-voice, plus workflow-specific plugins for ITSM or HR OOTB agents. Speak with your account team for full entitlement details.
Not yet. In the initial release, AI Voice Agents operate separately with a voice-optimized runtime. They can use many of the same tools (record operations, subflows, scripts, RAG search, MCP server), but the voice runtime is optimized for real-time interactions. The longer-term direction is to converge these runtimes into a single AI Agent framework powering both voice and chat.
Yes. Two common patterns: (1) Keep the CCaaS provider’s IVR tree for a global line with multiple departments, and route a specific option (e.g., “Press 1 for password reset”) to ServiceNow Voice. (2) For a dedicated IT or HR line, remove the IVR tree entirely and route directly to Now Assist in Voice.
Yes. KB Summarization and KB Search are available through the out-of-the-box Search Retrieval Tool. Tip: Filter to specific knowledge bases addressing common voice channel use cases rather than searching all KBs, to optimize response time.
An AI Voice Agent Analytics dashboard (accessible from the Analytics tab in Assistant Designer) aggregates key metrics including total voice conversations, deflected conversations, conversation outcomes, satisfaction scores, and AI agent performance. Post-call sentiment analysis calculates a 1–5 satisfaction score by analyzing the transcript. Real-time sentiment detection is being explored for future versions.
Yes. You can use a script tool call with a custom script to retrieve external information, or use the MCP Server tool to connect to other agentic systems. External Content Connectors are also supported and accessed using the Search tool.
Key roadmap items include: AI Voice Agent Testing Experience (Q1 ’26), mobile app channel support (Q1 ’26), outbound calling (Q2 ’26), GCC/regulated market support (Q1–Q2 ’26), ambient/filler sounds (H1 2026), additional CCaaS providers (3CLogic, Amazon Connect, Five9, NICE), and additional Tier 1 & 2 languages.
Industry Voice Agent Examples
The following examples illustrate how AI Voice Agents transform common industry scenarios — comparing the old NLU-based IVR approach with the new LLM-powered voice agent approach. Each example shows a routine use case and a complex use case to highlight the differentiation.
Notice how the AI Voice Agent doesn’t just route — it reasons, retrieves context, takes action across multiple systems, and resolves end-to-end in a single conversation.
Appointment Scheduling & Rescheduling
“Say scheduling or press 2” → Transfers to scheduling queue → Agent manually looks up patient record → Checks provider availability in separate EHR → Books appointment. Average: 8–12 minutes.
Caller: “I need to reschedule my Thursday appointment” → Agent authenticates via phone number + Soft PIN → Retrieves patient record & upcoming appointments → Checks provider availability via subflow → Confirms new slot → Updates record & sends SMS confirmation. Average: 2–3 minutes.
Post-Discharge Follow-Up with Medication Reconciliation
IVR cannot handle this scenario. Patient calls → Navigates 3–4 menu layers → Reaches nurse line → Nurse manually reviews discharge summary, pharmacy records, and care plan in separate systems → Escalates to pharmacist if mismatch found. Average: 20+ minutes, often requires callback.
Patient: “I just got home from the hospital and I’m confused about my new medications” → Agent authenticates → Retrieves discharge case, pulls medication list from care plan via RAG search → Cross-references active prescriptions → Identifies potential conflict → Summarizes medications in plain language → Creates escalation case for pharmacist review with full context → Sends follow-up summary via SMS. Single call, fully documented.
Claims Status Inquiry
“Say claims” → “Enter your policy number” → Reads back generic status (“Your claim is in review”) with no detail → “Press 0 for an agent” for follow-up questions. No context passed to agent on transfer.
Caller: “What’s happening with my auto claim from last week?” → Authenticates via phone + TOTP → Retrieves claim record with full timeline → Explains current stage, pending actions, expected dates → Offers to connect with adjuster if needed, passing full conversation context.
Multi-Line Policy Change with Compliance Verification
IVR can route to “policy changes” queue but cannot handle cross-policy logic. Agent must manually open each policy, check state-level regulations, calculate premium impacts, and verify beneficiary changes — often across 2–3 separate systems. Average: 25–35 minutes.
Caller: “I just moved to Texas and need to update my home and auto policies, plus change the beneficiary on my life policy” → Agent authenticates → Retrieves all active policies → RAG search for TX regulatory requirements → Calculates premium changes per policy → Validates beneficiary change requirements → Creates case for each policy change with pre-populated compliance fields → Transfers to licensed agent for final approval with full context. Single call, multi-policy.
Credit Card Transaction Dispute
“Say dispute” → “Enter last 4 digits of card” → Routes to disputes queue → Agent manually reviews transaction history, fills dispute form, initiates provisional credit. Average: 15+ minutes.
Caller: “I see a $247 charge from a restaurant I never visited” → Authenticates via phone + Okta Push → Retrieves recent transactions, identifies the specific charge → Collects dispute reason and details conversationally → Creates dispute case with pre-filled data → Initiates provisional credit workflow → Confirms timeline and sends reference number via SMS. Average: 3–4 minutes.
Fraud Alert Triage & Account Lockdown
Customer receives fraud alert SMS, calls in → “Press 1 for fraud” → Queue wait → Agent manually reviews flagged transactions one by one, locks card via separate fraud system, initiates replacement. Requires multiple holds and transfers. Average: 30+ minutes.
Caller: “I just got a fraud alert, I need help now” → Multi-factor auth (phone + TOTP) → Pulls all flagged transactions → Walks through each conversationally: “Was the $89 at Amazon on March 15th you?” → Marks confirmed/disputed for each → Initiates card freeze via API → Triggers replacement card workflow → Creates comprehensive fraud case with all transaction dispositions → Escalates to fraud team with full context. Single call, full resolution initiated.
Billing Inquiry & Payment Arrangement
“Say billing” → Reads total balance → “For payment arrangements, press 3” → Transfers to billing agent who re-verifies identity and re-pulls account. No context carried over. Average: 10+ minutes.
Caller: “My bill seems higher than usual and I need to set up a payment plan” → Authenticates → Retrieves current bill, compares to prior 3 months → Explains the variance (“Your bill is $34 higher due to international roaming charges on March 3–5”) → Offers payment arrangement options → Sets up 2-installment plan → Confirms via SMS. Average: 3 minutes.
Service Outage Triage with Proactive Network Correlation
Customer calls about outage → IVR plays generic outage message → If no match, routes to tech support queue → Agent manually checks network status tool, runs diagnostics, creates ticket. No correlation to known outages. Average: 15–20 minutes.
Caller: “My internet has been down for 2 hours” → Authenticates → Correlates caller’s service address with active network incidents via CMDB → Finds known fiber cut in area → Provides specific ETA from network ops team → Offers to set up proactive SMS notifications when service is restored → If no known outage, runs remote line diagnostics via subflow → Either resolves or creates prioritized ticket with full diagnostic data for field dispatch.
License Management & Access Provisioning
“Say licensing” → Routes to support → Agent verifies entitlements in CRM, then navigates to provisioning system to add user. Often requires escalation to license admin. Average: 12+ minutes.
Caller: “I need to add 5 users to our enterprise license” → Authenticates as account admin → Checks current license count and available seats → Confirms request details → Provisions users via catalog item execution → Sends activation emails → Updates account record. Average: 2 minutes.
P1 Production Incident with Multi-System Impact
Customer calls for urgent production issue → “Press 1 for support” → Queue wait → L1 agent collects info, creates ticket, escalates to L2 → L2 manually checks monitoring tools, CMDB, change records → Conference calls engineering. Multiple handoffs, repeated context. Average: 45+ minutes to engage right team.
Caller: “Our production environment is down, this is critical” → Authenticates → Checks for known incidents and recent changes via CMDB → Finds recent deployment change record matching timeframe → Creates P1 incident with auto-populated CI, change correlation, and customer impact statement → Triggers escalation workflow to on-call engineering team → Subscribes caller to live status updates → Provides incident number and expected response SLA. Average: 5 minutes to full engagement.
Permit Application Status & Document Submission
“Press 3 for permits” → Routes to permits office during business hours only → Agent manually searches permit database → Reads status. Document submission requires in-person visit or mail. Average: 10+ minutes (if call is answered).
Caller: “I applied for a building permit 3 weeks ago, what’s the status?” → Authenticates via phone + knowledge factor → Retrieves permit application with full review timeline → Identifies missing document holding up review → Sends SMS link to upload document via portal → Updates case with notification of outreach. Available 24/7. Average: 3 minutes.
Multi-Agency Benefits Eligibility & Enrollment
Citizen calls about benefit eligibility → IVR routes to one specific program → Agent can only check that program’s eligibility → Citizen must call separate numbers for housing, food assistance, healthcare. Each call: 15–20 minutes. Total: 3–4 calls over multiple days.
Caller: “I recently lost my job and need to know what help I can get” → Authenticates → Collects household details conversationally → Checks eligibility across multiple programs (unemployment, SNAP, Medicaid, housing assistance) via API integrations → Summarizes all eligible programs → Initiates enrollment for consented programs → Creates unified case with cross-agency references → Sends comprehensive summary via SMS/email. Single call, all programs.
Key Takeaways
Explore More
Continue exploring ServiceNow Contact Center capabilities
