Poly

PolyAI develops an "Agentic Dialog Platform" that enterprises use to build, run, adapt and govern AI dialog agents handling customer voice and chat conversations. Rather than deflecting or routing, the agents are positioned to complete tasks end to end — bookings and reservations, secure payments and refunds, ID verification, claims and account servicing, troubleshooting, and escalation to human agents with full conversation context. Agents run across voice and messaging channels and support language switching mid-conversation; the company cites 45 languages as standard on its marketing site and 24+ languages in its developer documentation, with sub-300ms response latency.

The stack is built on proprietary components: Raven, a voice-native LLM trained on more than one billion real enterprise conversations, an in-house ASR system called Owl designed for accents, background noise and interruptions, and text-to-speech blending human recordings with neural synthesis. Agent behavior is shaped across four areas — behavior/guardrails, grounded knowledge sources, multi-step flows, and voice — with changes tested in a shareable pre-production environment. Builders can work through Agent Studio (no-code/low-code), an ADK for local YAML/Python development and CLI deployment, or REST APIs; a natural-language agent-drafting feature called Wren is also offered. Governance features include SOC 2, HIPAA, GDPR and PCI DSS compliance as standard and visibility into agent decisions.

Deployments integrate with contact center stacks via SIP or PSTN into a CCaaS platform or telephony provider, with other systems connected through APIs, telephony, CRM, payments and MCP connectors. As of December 2025 the company reported 100+ enterprise customers, over 2,000 live deployments in more than 25 countries, and said its agents perform the work of 1,000+ full-time employees at multiple enterprises.

Founding story

PolyAI was founded in 2017 by Cambridge machine learning researchers on the premise that enterprises should not have to choose between scale and quality in customer service and that voice AI could handle genuinely complex dialog rather than only simple requests. Forbes describes the company as a University of Cambridge spinout; CEO Nikola Mrkšić previously worked with Apple on Siri. Steve Young, creator of the HTK speech toolkit and a University of Cambridge Pro-Vice-Chancellor, is an advisor.

Business model

B2B enterprise software: PolyAI sells a platform on which enterprises deploy voice and chat dialog agents for customer service, with implementation typically cited at about six weeks from kickoff to launch. The platform is offered through multiple build surfaces (Agent Studio, ADK, REST APIs) and, per a 2025 announcement, was opened to all builders with an initial free tier.

Sources describe enterprise contracts and recurring revenue but do not detail pricing. Forbes reported GAAP revenue of $15 million in the 12 months to 31 January 2025 (up from $8.9 million a year earlier) and management expectation of ARR doubling to $40 million.

Traction

As of December 2025: 100+ enterprise customers, 2,000+ live deployments, 45 languages, 25+ countries; agents doing the work of 1,000+ full-time employees at multiple enterprises and roughly $1 billion in annual value created across the customer base by the company's estimate. Forrester's Total Economic Impact study reports 391% ROI and average savings of $10.3 million. Forbes reported GAAP revenue rising to $15 million (FY to Jan 2025) from $8.9 million, ARR expected to reach $40 million, US client revenue nearly tripling, and losses widening to $26.6 million from $23.8 million. Customer-cited outcomes include Fogo de Chão projecting just over $7M in incremental revenue and 95% customer satisfaction, and Melting Pot generating $300k from after-hours bookings.

Latest developments

December 2025: $86 million Series D co-led by Georgian, Hedosophia and Khosla Ventures at a $750 million valuation, with funds directed at Agent Studio technology and go-to-market expansion. The Agentic Dialog Platform was opened to all builders, a Canadian technology hub and senior leadership hires were announced, and documentation (last modified August 2026) introduces Wren for natural-language agent drafting alongside Agent Studio, the ADK and REST APIs.