PolyAI

Nikola Mrkšić was born in Serbia and grew up a mathematics prodigy who won a scholarship to study computer science at Cambridge. Before co-founding PolyAI, he had worked on Apple's Siri voice assistant as a machine learning researcher. He met his co-founders Tsung-Hsien Wen and Pei-Hao Su at Cambridge's Machine Intelligence Laboratory, where all three were researching dialogue systems. They founded PolyAI in London in 2017 as a spinout from Cambridge, with a straightforward observation: every large enterprise call centre was fronted by a voice assistant that customers instantly tried to circumvent, and nobody had solved that problem.

Consumer voice assistants — Siri, Alexa, Google Assistant — were moderately useful but not essential. If removed from a device, most users would barely notice. Enterprise call centres were different: every missed call or frustrated caller represented a direct business cost. Staff churn at contact centres regularly exceeded 200% annually because of poor pay and difficult conditions. The solution PolyAI built was not a cheaper version of what already existed — it was a voice agent trained specifically on the complexity of real customer conversations, capable of handling topic changes, ambiguous requests, and accented speech, in 45 languages.

PolyAI's key differentiation was that it developed its own large language models rather than plugging into OpenAI or ElevenLabs voice technology, giving it an edge in realism and control. The company built its customer base in hospitality, financial services, logistics, and energy: FedEx, Marriott, Caesars Entertainment, UniCredit, and PG&E all deployed PolyAI agents to handle millions of calls. Revenue grew from $8.9 million in the fiscal year ending January 2024 to $15 million in the following year. The company raised a $50 million Series C in May 2024 at a $500 million valuation.

In April 2025, PolyAI launched Agent Studio — a platform giving enterprises visibility into how their voice agents behaved, why they responded as they did, and how they could be tuned over time. The platform emphasised transparency and governance for regulated industries where a hallucinating call centre agent was a liability risk rather than merely an inconvenience. By 2025, PolyAI agents were handling customer calls about power outages for Pacific Gas & Electric, lost credit cards for UniCredit, and restaurant bookings for Las Vegas casinos. The company's AI agents were doing the equivalent work of more than 1,000 full-time employees at multiple enterprises, creating approximately $1 billion in total economic value annually across its customer base.

In December 2025, PolyAI raised $86 million in a Series D led by Georgian, Hedosophia, and Khosla Ventures, with NVIDIA Ventures, Citi Ventures, and Zendesk Ventures also participating. The round valued the company at $750 million, a 50% increase from five months earlier. ARR had doubled in the year to over $40 million, with US revenue nearly tripling. PolyAI served 100+ enterprise customers across 2,000 live deployments in 45 languages and 25 countries. A Forrester Total Economic Impact study found PolyAI customers achieved 391% ROI with average savings of $10.3 million per enterprise. The Financial Times ranked PolyAI as Europe's fastest-growing enterprise AI company.