Salesforce adds voice to Agentforce: What support leaders need to know
Salesforce is rolling out Agentforce Voice, letting AI agents answer inbound calls, not just chats. The move lands ahead of Dreamforce and extends Agentforce's footprint, which has already passed 12,000 implementations.
For customer support teams, this means AI can now sit at the same telephony layer as your agents-handling greetings, routing, basic troubleshooting, and handing off when it hits a wall.
What Agentforce Voice does
- Handles live phone conversations with natural back-and-forth. Callers can interrupt mid-sentence.
- Lets you set tone, speed, and pronunciation, so brand and product names sound right.
- Integrates with major contact center and telephony stacks: Amazon, Five9, Genesys, Nice, and Vonage (Ericsson).
Salesforce will also offer early access in November to Agent Script-tooling to control what agents say and do across scenarios. That matters for compliance, consistency, and faster iteration.
Why this matters for support
- Deflection and containment: Offload repetitive Tier 1 calls (password resets, order status, simple troubleshooting).
- Speed: Shorter handle times with structured call flows and real-time lookups.
- Consistency: Scripted answers reduce variance while keeping a natural voice.
- Brand control: Dial in tone and pacing to match your service standards.
Market backdrop and pressure
Voice is moving fast across AI. OpenAI and Anthropic both enabled smooth spoken conversations, and that capability is now landing inside business software.
Salesforce faces questions on AI impact and adoption pace. The company noted AI could shift workforce needs and demand, while leadership said change will be incremental, not a wipeout. Analysts at RBC said enthusiasm around Agentforce cooled as adoption lagged expectations.
Meanwhile, competition is heating up. Former Salesforce co-CEO Bret Taylor's startup Sierra says its AI agents "can now pick up the phone," with clients like ADT, SiriusXM, and SoFi. On Wall Street, Salesforce shares are down about 28% in 2025, while the Nasdaq is up roughly 15%.
Where this fits in your stack
- Telephony: Plug into Amazon, Five9, Genesys, Nice, or Vonage to route calls to AI first, then to humans when needed.
- Knowledge and CRM: Performance will track the quality of your knowledge base, CRM data, and policy rules.
- Escalations: Plan clean handoffs with full context in the agent's console to avoid caller repetition.
Limits and risks to plan for
- Edge cases: Complex billing, legal, or safety issues should escalate fast.
- Latency: Voice AI needs tight response times-test on real traffic before scaling.
- Compliance: Log consent, disclosures, and redaction for PII/PHI.
- Quality: Test accents, noisy environments, and domain-specific terms.
30-day pilot plan
- Pick 3-5 high-volume intents (e.g., order status, subscription changes, password reset).
- Write Agent Script variants per intent: greeting, verification, answer, fallback, escalation.
- Tune voice settings: tone, speed, and key pronunciations (product names, acronyms).
- Route 5-10% of eligible calls to AI during business hours for two weeks; expand if metrics hold.
- Daily review: call recordings, failure reasons, and handoff quality. Ship script updates every 24-48 hours.
Metrics to track
- Containment rate (resolved without human)
- Average handle time (AI and human)
- CSAT/CES on AI-handled calls
- Escalation rate and reasons
- First contact resolution and reopen rate
- Cost per contact vs. baseline
Questions to ask Salesforce (or any vendor)
- What is the average and p95 latency for voice responses?
- How do you manage barge-in, interruptions, and turn-taking?
- What tools exist for script testing, versioning, and rollback?
- How do you handle redaction, retention, and audit trails for call recordings?
- What's the escalation protocol and data handoff into the agent desktop?
- Which languages and accents are production-ready today?
Context and useful references
Skill up your team
If you're standing up an AI-assisted contact center, upskilling agents and managers on prompt strategy, escalation design, and QA will shorten your ramp.
- AI courses by job role for support leaders and frontline agents
The bottom line
Agentforce Voice moves phone support into AI territory. Start small, script tightly, measure daily, and expand where the data proves it.
Dreamforce will bring more details, but you can begin with a focused pilot and a clear metric target. Keep humans in the loop, and let AI handle the repetitive load.
Your membership also unlocks:
