elevenlabs did this differently
guide on how you can too
👋 Hey, I’m Suhas, and welcome to another edition of TPF Weekly!
Quick poll before we get into it:
Three weeks after launching their film-dubbing startup in January 2023, Piotr Dąbkowski and Mati Staniszewski observed a unique user pattern.
Almost nobody was using it for film dubbing.
Instead, the YouTubers showed up, mainly wanting plain text-to-speech for their videos.
One early audiobook author even hit their beta 500 times to bypass platform AI detectors and ship a full novel.
They had two choices: insist on the original framing (the one their investors had backed) or follow the users.
They chose the latter.
ElevenLabs has never really built a roadmap in the conventional sense.
They watched what their users were already doing, then shipped it as the next product layer.
Dubbing app that became a text-to-speech company
Piotr and Mati had thought film dubbing was the company.
The pitch was to keep the original actor’s voice and emotion across languages, which would solve a real problem in markets like Poland, where television uses a convention called lektor (one male narrator reads every character’s lines, flat and unemotional, over the original audio).
The markets they got instead, were different.
So the team reshaped the roadmap around the API users were hitting.
AI Dubbing launched later that year as a parallel product, and it remains one of ElevenLabs’ core offerings today.
This is the move I keep coming back to as a PM lesson.
Founder stories usually run on “we saw the future and built it”.
ElevenLabs’ story is closer to: we watched what our customers were grinding through, and we built the thing they kept rebuilding themselves.
API that got pulled into an agents platform
By mid-2024, ElevenLabs had multilingual TTS, voice cloning, a voice library, long-form Projects, and dubbing.
The obvious next step in any API startup playbook is to deepen the existing surface by adding more models, lower latency, cut prices, ship more SDKs.
They did some of that but the bigger move was Conversational AI (later rebranded as ElevenAgents), which launched in November 2024.
Teams trying to ship a voice agent in 2024 were running into the same wall.
Stitching together speech-to-text, an LLM, text-to-speech, telephony, and an orchestration layer meant managing 5 vendors, SLAs, latency budgets, and places where the conversation could fail.
Most of that engineering effort had nothing to do with the product the team was trying to build.
ElevenLabs’ bet was that owning the full voice stack natively would mean lower latency, fewer failure points, and better turn-taking.
The moat is in deployments where stitched-together stacks tend to break down, like Indian cities with noisy backgrounds, Hinglish conversations, or low-end Android devices.
If you want to see it for yourself:
The numbers since then suggest the bet paid off as the platform has hosted over 2 million agents handling 33M+ conversations to date.
Cars24 is running 3 million+ minutes of AI-supported calls across 300+ retail hubs in India, with a 50% reduction in calling costs and 45% of sales now supported by AI-assisted calls.
Meesho is handling 60,000 customer calls a day in Hindi and English.
Razorpay is running outbound Hinglish voice agents that have completed 35,000+ calls with a ~28% connection rate, on par with human call center benchmarks.
Look at the revenue curve
That last number was $330M ARR at the end of 2025, and crossed $500M ARR in the first four months of 2026.
Twilio, the closest historical comparison, took about 8 years to hit the same $330M.
The easy read on this is “ride the AI wave” but that doesn’t survive the data.
OpenAI shipped a Realtime API in the same window ElevenLabs shipped Agents, while Cartesia and Hume each shipped credible TTS competitors at lower latency.
Voice AI was 22% of the most recent YC class.
If the compounding were just market gravity, every voice startup would be at 9-figure ARR.
ElevenLabs had something rare in the form of a customer base already building the next product on top of the previous one, which meant each new layer launched into demand that already existed.
What this means for your roadmap
Most product teams I work with show an inverse relationship between roadmap confidence and user evidence.
The bets they’re most certain about have the least user evidence underneath.
Tickets and observed behaviour anchor the next quarter; nothing much anchors the year after that.
ElevenLabs flipped that.
Their 18-month bets were customer behavior they had already observed at scale.
The TTS pivot, Projects, Conversational AI (now ElevenAgents), the Iconic Voice Marketplace (where users were already trying to license celebrity voices off-platform) all followed the same shape.
The strategy has one prerequisite though: you have to be close enough to see what users are building.
You have to breathe support tickets, live in discord channels, sales calls, and drown in the inbound your sales team logs.
Mati still interviews every hire at 500+ headcount, which has less to do with hiring than with staying one degree of separation from the company’s signal.
The harder part is that when you spot what users are already doing, you have to kill the roadmap you already sold internally.
ElevenLabs reshaped the dubbing-first plan to ship raw TTS. Two years later they dropped the “deepen the API” playbook to make room for Agents.
Both times they walked away from a plan they had built consensus around.
Now, look at your current roadmap.
The items you’re most confident about are the ones with the least user evidence underneath them.
Worth a second look this week.
Upcoming Events
Vibe Sprint | Delhi NCR
23 May
Register Here
AI Mixer | Mumbai
29 May
Register Here
Vibe Sprint | Bengaluru
30 May
Register Here
GrabChai | Chennai
30 May
Register Here
Exclusive Jobs of the Week
Senior Product Manager
Lead Product Manager
Director Product
These and other roles open across top companies like Meesho, Google, Zomato & many more.
Reply with the weirdest thing you’ve ever seen users do with your product.
I read every one.
Cheers,
Suhas 👋🏻
P.S. This gets better when the right people are in the room. Share it with one.