PolyAI
Enterprise voice AI agents that answer and resolve customer calls
Last reviewed 2026-06-19
PolyAI builds enterprise voice AI agents that answer inbound customer calls in natural language, handle interruptions, complete transactions via backend function calling, and hand off to humans with full context when needed. A Cambridge spinout, it is positioned as an agentic dialog platform with two build paths: a no-code Agent Builder and a developer ADK with IDE, Git versioning, and CI/CD, both running on a shared runtime. It supports voice plus chat, SMS, and WhatsApp. Founded in 2017 by Nikola Mrksic, Pei-Hao Su, and Tsung-Hsien Wen, PolyAI serves large contact centers in hospitality, gaming, banking, retail, healthcare, and telecom. It runs a proprietary model stack ("Raven") trained on enterprise conversations rather than wrapping a single third-party LLM. Call containment is reported above 50%, with the remainder escalating, so in practice it is a supervised agent rather than a full human replacement.
What it can do
Answer and resolve inbound voice calls
SupervisedHandles natural-language calls with interruptions and resolves common requests; containment is reported above 50%, with the rest escalated.
sourceExecute backend actions via function calling
SupervisedCalls backend systems to authenticate callers, process refunds, and look up records during a conversation.
sourceHand off to a human with full context
SupervisedPerforms a warm handoff to a human agent with full conversation context and annotations.
sourceBuild and version agents (ADK)
AssistantA developer kit provides an IDE, Git-based versioning, and CI/CD for building and shipping agents.
source
Strengths
- +Strong, natural voice quality proven on hard and regulated call types
- +Deep CCaaS and CRM integration plus a proprietary model stack with enterprise compliance
- +Real enterprise traction and NVIDIA backing
Limitations
- −Opaque, high-floor pricing inaccessible to SMBs
- −Historically heavy managed-service implementation; the self-serve story is newer
- −Containment above 50% means a large share of calls still escalate
Overview
PolyAI builds enterprise voice AI agents that answer inbound customer calls in natural language. A Cambridge spinout, it is positioned as an agentic dialog platform with a no-code Agent Builder and a developer ADK on a shared runtime.
What it does
Agents answer calls, handle interruptions, complete transactions via backend function calling (authentication, refunds, lookups), and perform a warm handoff to human agents with full context when needed. It also supports chat, SMS, and WhatsApp. Reported call containment is above 50%, with the remainder escalating.
Integrations & setup
Connects to telephony and CCaaS platforms (Five9, NICE CXone, Twilio, Amazon Connect, Genesys), CRMs (Salesforce, Zendesk, HubSpot, Dynamics 365), and vertical systems such as Epic in healthcare. It supports function calling and MCP for tool discovery, and runs a proprietary model stack ("Raven").
Pricing
Usage-based, per-minute, with custom annual enterprise contracts. No public starting price; third-party estimates exist but are not vendor-confirmed.
Best for / not for
Best for large contact centers that need natural, transactional voice automation with deep integration. Not a fit for SMBs needing self-serve, low-cost tooling.
Traction
PolyAI has raised over $200M across rounds, including a reported $86M Series D in December 2025 at a $750M valuation, with backers including NVIDIA, Georgian, Hedosophia, and Khosla Ventures. It serves 100+ enterprise customers across hospitality, gaming, banking, retail, and healthcare.
Alternatives
Parloa competes in enterprise voice contact-center automation; Vapi, Bland AI, and Retell AI target the developer/voice-infrastructure end.
What people are saying
We aggregate real LinkedIn discussion into sentiment for the agents people search most. PolyAI isn't tracked yet, want it added? Request tracking.
FAQ
Is PolyAI fully autonomous?+
No. It resolves a share of calls autonomously (reported containment above 50%) and performs a warm handoff to humans for the rest. In practice it is a supervised agent.
What model does PolyAI use?+
PolyAI runs a proprietary model stack ("Raven") trained on enterprise conversations rather than wrapping a single third-party LLM.
Sources
- PolyAI (official site) · accessed 2026-06-19
- PolyAI documentation · accessed 2026-06-19
- PolyAI for developers · accessed 2026-06-19
- PolyAI raises $86M Series D (PRNewswire) · accessed 2026-06-19
Last reviewed 2026-06-19