navyapandir
ServiceNow Employee

ServiceNow AI Voice Agents

GenAI-powered voice agents built for natural, dynamic phone conversations that resolve issues in real time

Now Assist Powered • CCaaS Integration Required for PSTN Calls • App calling supported without CCaaS 

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.

45-60%

of customers choose voice even when digital options exist

#1

channel for urgent, complex, and high-emotion interactions

73%

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:

DTMF IVR
1990s–2010s

“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.

NLU-Based IVR
2015–2023

“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.

AI Voice Agents
2024+

“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

Natural, Multilingual Conversations

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.

CCaaS Platform Integration

Seamless integration with Genesys Cloud and Twilio at launch, with 3CLogic, Amazon Connect, Five9, and NICE support planned for future releases.

Graceful Fallback & Escalation

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.

Voice Personality Library

Personalize your agent with an extensive voice library. Configure persona, language, welcome message, and tone to match your brand.

Multi-LLM Support

Supports Azure OpenAI, Google Gemini, and Anthropic Claude on AWS — accessed via ServiceNow-managed private connections for security and performance.

Analytics & Transcripts

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.

OOTB Voice Agents (Shipping Q4’25)

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
Upcoming Use Cases (Q1’26+)
  • 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:

1
Caller Speaks
Phone / WebRTC / Mobile
2
Speech-to-Text
Whisper Large 3
3
LLM Reasoning
Intent + Context + Memory
4
Tool Execution
Flows, APIs, KB Search
5
Text-to-Speech
Cartesia Sonic 2

Key Components

Voice Orchestrator

Hosted in the ServiceNow Compute Hub. Manages the full call lifecycle including routing, turn-taking, agent selection, authentication, context gathering, and tool orchestration.

LLM Engine

Azure OpenAI, Anthropic Claude on AWS, and Google Gemini for reasoning. Accessed via ServiceNow-managed private connections to provider-hosted infrastructure.

Tool Framework

Flows, subflows, scripts, search retrieval, knowledge graphs, MCP server tools, and record operations — all executing within the customer instance with ACL enforcement.

Memory

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.

Voice Assistant (Assistant Designer)

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.

Voice Agent (AI Agent Studio)

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:

Caller Authentication
  • Phone number identification
  • Soft PIN (6-digit enrolled)
  • TOTP (Authenticator app)
  • SMS OTP (via Twilio)
  • Push Notification (Okta Verify)
  • Multi-factor support
Content Guardrails
  • 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
Data Privacy
  • 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
Access & Evaluation
  • 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.

Launch Voice Agent Directly

Users tap a Prominent Action Button on the mobile home screen to immediately start a voice conversation.

Launch from Now Assist Virtual Agent

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

Assist Consumption

50 Assists per call. Calls shorter than 30 seconds don’t count toward consumption.

CCaaS Requirement

Genesys Cloud and Twilio supported at launch. Additional providers coming in Q1 2026. Omni-channel starting with Mobile App in Q1 2026.

Different from Standard AI Agents

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.

Multi-lingual Support

English, German, French, and Spanish supported. Configured at the Voice Assistant level. Additional Tier 1 and Tier 2 languages coming soon.

Current Deployment Constraints (Q4 Release):

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:

Key Implementation Steps
1
Prerequisites & Planning

Ensure you have Now Assist licensing, a supported CCaaS provider (Genesys Cloud or Twilio), and the AI Voice Agents plugin activated on your instance.

2
Configure Voice Assistant

Use Assistant Designer to set up voice personality, authentication method, safeguard behaviors, and CCaaS provider connection.

3
Create or Select Voice Agents

Use OOTB agents for ITSM, HR, or CSM, or build custom agents in AI Agent Studio with natural language instructions, tools, and access rules.

4
Map Agents & Configure Tools

Assign Voice Agents to the Voice Assistant. Map required tools (flows, knowledge search, record operations) to each agent.

5
Test with Agent Evals

Run simulations using golden truth data (up to 1,000 scenarios) before production. Review analytics dashboard for performance tuning.

6
Activate & Deploy

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

What advantages do AI Voice Agents offer over CCaaS-built voice bots?

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.

What license and patch level is needed?

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.

Do Voice Agents share the same runtime as chat-based AI Agents?

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.

Can AI Voice Agents be activated within an existing IVR call tree?

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.

Will KB articles be summarized and read back during voice calls?

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.

What analytics and sentiment capabilities are available?

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.

Can AIVA access external data outside ServiceNow?

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.

What’s on the roadmap?

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.

🏥 Healthcare
ROUTINE USE CASE

Appointment Scheduling & Rescheduling

OLD: NLU-Based IVR

“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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (patient lookup) • Subflow (provider availability check) • Flow Action (appointment update) • Script (SMS confirmation via Twilio)
COMPLEX USE CASE

Post-Discharge Follow-Up with Medication Reconciliation

OLD: NLU-Based IVR

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.

NEW: AI Voice Agent

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.

Tools Used: Search Retrieval / RAG (care plan, KB) • Record Operation (discharge case, medication records) • Now Assist Skill (summarize for patient) • Flow Action (create escalation case) • Script (SMS follow-up)
🏦 Insurance & Financial Services
ROUTINE USE CASE

Claims Status Inquiry

OLD: NLU-Based IVR

“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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (claim lookup, timeline) • Search Retrieval (policy KB) • Now Assist Skill (summarize claim status) • CSR Handoff (with context)
COMPLEX USE CASE

Multi-Line Policy Change with Compliance Verification

OLD: NLU-Based IVR

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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (multi-policy lookup) • Search Retrieval / RAG (state compliance rules) • Script (premium recalculation via external API) • Flow Action (create multiple cases) • CSR Handoff (licensed agent with context)
💳 Banking
ROUTINE USE CASE

Credit Card Transaction Dispute

OLD: NLU-Based IVR

“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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (transaction lookup) • Flow Action (create dispute case, initiate provisional credit) • Script (SMS confirmation) • Subflow (compliance tagging)
COMPLEX USE CASE

Fraud Alert Triage & Account Lockdown

OLD: NLU-Based IVR

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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (flagged transactions) • Script (card freeze API, replacement card workflow) • Flow Action (create fraud case with dispositions) • MCP Server (connect to external fraud system) • CSR Handoff (fraud specialist with full context)
📱 Telecommunications
ROUTINE USE CASE

Billing Inquiry & Payment Arrangement

OLD: NLU-Based IVR

“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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (billing history, account lookup) • Now Assist Skill (bill comparison summary) • Subflow (payment arrangement creation) • Script (SMS confirmation)
COMPLEX USE CASE

Service Outage Triage with Proactive Network Correlation

OLD: NLU-Based IVR

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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (service address, CMDB) • Search Retrieval (active network incidents) • Subflow (remote line diagnostics) • Flow Action (create incident, subscribe to notifications) • MCP Server (network monitoring system)
💻 Technology / SaaS
ROUTINE USE CASE

License Management & Access Provisioning

OLD: NLU-Based IVR

“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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (license/account lookup) • Catalog Item (user provisioning) • Flow Action (send activation emails) • Script (update account record)
COMPLEX USE CASE

P1 Production Incident with Multi-System Impact

OLD: NLU-Based IVR

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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (CMDB, change records, active incidents) • Search Retrieval (recent deployments) • Flow Action (create P1 incident, trigger escalation) • Subflow (on-call team notification) • Script (status subscription setup)
🏛 Public Sector & Government
ROUTINE USE CASE

Permit Application Status & Document Submission

OLD: NLU-Based IVR

“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).

NEW: AI Voice Agent

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.

Tools Used: Record Operation (permit application lookup) • Search Retrieval (requirements checklist) • Script (SMS with upload link) • Flow Action (update case with outreach record)
COMPLEX USE CASE

Multi-Agency Benefits Eligibility & Enrollment

OLD: NLU-Based IVR

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.

NEW: AI Voice Agent

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.

Tools Used: Record Operation (citizen profile, case creation) • MCP Server / Script (multi-agency eligibility APIs) • Search Retrieval / RAG (program requirements KB) • Flow Action (enrollment initiation per program) • Now Assist Skill (summarize eligibility results) • Script (SMS/email summary)

Key Takeaways

1. Voice agents resolve — they don’t just route. Natural conversation replaces IVR trees with real-time intent understanding and action.
2. Everything stays on-platform. Data, workflows, knowledge, and intelligence — no external extraction, no latency, no security gaps.
3. CRM-ready OOTB use cases: Check Case Status, Create/Update Case, Order Operations — plus Build Your Own with natural language.
4. Security is built in: caller auth (4+ methods), Now Assist Guardian (16 categories), PII masking, prompt injection protection, Agent Evals.
5. Start small, iterate fast, measure business outcomes. Run Agent Evals before production. Your first deployment won’t be perfect — and that’s okay.

Explore More

Continue exploring ServiceNow Contact Center capabilities

Version history
Last update:
5 hours ago
Updated by:
Contributors