Mati Staniszewski on ElevenLabs, Voice AI & Building the Communication Layer for AI

What It Covers

ElevenLabs co-founder Mati Staniszewski explains how he and co-founder Piotr built a focused audio AI research-and-product company from 2022, starting with dubbing and evolving into a platform serving voice agents, marketing, and accessibility across industries including FinTech, healthcare, and telecom, with Deutsche Telekom as a deeply integrated enterprise customer.

Key Questions Answered

• •Focused Research Defense: ElevenLabs deliberately restricts its research team to audio models only — text-to-speech, speech-to-text, transcription, and orchestration — as a competitive moat against larger labs. Before pursuing any new product, the internal test is whether audio represents the primary bottleneck. If a product's core problem is video or image quality, ElevenLabs explicitly passes, as demonstrated when they declined to build avatar and lip-sync tools two years ago.
• •FDE Model from Palantir: Forward-deployed engineers (FDEs) sit inside the product team, not go-to-market, at ElevenLabs. Teams of roughly five engineers embed directly with enterprise clients — physically traveling to their offices — to build integrations with CRM systems, SIP trunking, and phone infrastructure. Critically, every integration insight gets abstracted back into the core platform, so learnings from one client automatically improve the product for thousands of others.
• •Voice Marketplace as Ecosystem Lock-in: ElevenLabs built a 20,000-voice marketplace where individuals create authenticated AI voices, then earn passive compensation each time their voice is used. This creates a proprietary data asset and switching cost that pure model quality cannot replicate. When model performance converges across competitors, the ecosystem — brand trust, voice library, and revenue-sharing structure — becomes the durable differentiator.
• •Customer Discovery Before Product: When ElevenLabs initially pitched dubbing to creators, feedback revealed more immediate needs: editing recorded lines, narrating scripts before filming, and replacing their own voice in videos. This redirected the entire research roadmap toward high-quality single-language text-to-speech before tackling multilingual dubbing. Founders should treat early customer conversations as product prioritization data, not just validation exercises.
• •Small Teams with Embedded Engineers Across All Functions: ElevenLabs caps teams at under 10 people and embeds engineers inside legal, talent, and operations — not just product. This structure enables bottom-up AI adoption without top-down mandates. The revenue operations function uses AI tools to auto-transcribe sales calls, pre-fill CRM fields, and run a voice-based AI SDR on the website that collects significantly more prospect information than traditional dropdown forms.
• •FinTech Leads Enterprise Voice Agent Adoption: Among ElevenLabs' enterprise verticals, FinTech — including Revolut, Klarna, and PagBank — moves fastest on deploying conversational voice agents, followed by healthcare and telecom. Retail and e-commerce began adopting at scale only in the current year. Enterprise expansion follows a prove-then-scale sequence: demonstrate measurable impact on one use case, complete deep integrations, run simulated call trials, then expand to additional product lines within the same account.