The core message of the session was that enterprise voice AI is moving from experimental demos into operational infrastructure: ElevenLabs wants companies to replace or supplement typed, menu-driven interactions with natural voice agents that can work across customer support, internal training, operations, marketing, and other workflows. Gordian Braun introduced ElevenLabs as a European Voice AI and audio model company, saying it has been in market for three and a half years, has 40 million users globally, is valued at 11 billion, and is trusted by about 80% of Fortune 500 companies. He framed the company’s role by comparison to OpenAI: what OpenAI is often seen as for text, ElevenLabs aims to be for audio and voice models.

Braun then demonstrated the company’s latest voice model, released the previous day, to show how far AI voices have moved away from robotic output. The demo emphasized emotional expression, speaker similarity, tone, texture, and intent, including professional voice clones that do more than imitate a voice. From there, he shifted into enterprise deployment, explaining that ElevenLabs’ models are typically used in agents that communicate with customers or internal users. These agents can support multiple channels such as phone, web interfaces, WhatsApp, email, Slack, and chat, and can switch languages within a conversation, including English, French, German, Turkish, and other languages needed in a region.

The practical implementation path he recommended starts with defining priority use cases and beginning smaller to build trust before scaling to larger deployments. He stressed that companies must prepare their data, define what success means, validate early tests, and then move toward production with iteration. Example use cases included customer support without waiting times, internal training and onboarding, delivery driver training, candidate screening, marketing voiceovers and dubbing, CEO videos in different languages, intake forms, vacation management, appointment booking, and customer support at scale. Braun also described the build process: structure enterprise data, connect RAG systems and APIs, design agents through a visual dashboard, add guardrails so sub-agents only access approved systems, deploy across preferred channels, and monitor success rates for future A/B tests and improvements.

In the technical architecture section, Braun explained that successful voice agents require multiple AI components working together: text-to-speech, transcription, validation, language detection, background-noise handling, turn-taking, system connections, and human escalation. He emphasized that AI should decide when to hand over to a person because “AI’s, they’re not perfect.” He closed the main presentation by highlighting deployment flexibility, European hosting, zero-retention modes, compliance with European requirements, and an AI UC one certification that made the company insurable. The Q&A began with a practical question about using voice to add tasks, calendar entries, and emails while dictating under the shower; Braun answered that this would be one of many possible agent use cases.