Two AI founders explain how voice and legal AI reached hundreds of millions in revenue.
It took roughly 20 months from the first text-to-speech release at the beginning of 2023 to the first $100 million in ARR, then about 10 months to $200 million and five months to $300 million, which closed out last year. Now the company is at $600 million.
Answered around 1:16600 people today, which is very quick growth on the people side, so the interview cycle, bringing people on board and onboarding are all deliberately optimized to keep the culture. From the first ten people there has been zero attrition; that core research and engineering group is all still there.
Answered around 2:26Three things. Everything generated is traced, so action can be taken when needed. Moderation now runs at both the voice and text level, so commercial or scam-like input gets flagged and can be blocked. And anyone can upload a sample to learn immediately whether it is AI, covering other open-source models too.
Answered around 19:03None, and it never did. The ideal person in that role could code, understand the customer and understand design, and very few people are expert in all of those at once, so the company hires experts in at least one field who understand at least one other really well.
Answered around 7:11Engineers are embedded everywhere, including non-engineering teams: the talent team has an engineer, legal has one, revenue and go-to-market engineering too. They create automations inside the team and make sure everyone adopts AI with a security check on everything deployed. Using too little of that software is wrong; using too much is also a flag.
Answered around 5:18Over $22 million has been paid back so far to the community of talent. People create their voice, it gets authenticated, they share it and they earn money. On price it works both ways: leave the default so ElevenLabs can distribute it more optimally, or pick your own, since use cases differ.
Answered around 21:04As a platform it offers all the models — Anthropic, OpenAI, open source, Google — so customers build their harness, agent orchestration and voice layer without depending on any one model. The defining focus is the single layer of interaction and communication, where it has outcompeted them on text-to-speech, speech-to-text, turn-taking and music.
Answered around 25:56A sustained 50% quarter over quarter for the last seven quarters. As of the close last week on Tuesday, it became one of the fastest enterprise companies with a direct sales motion to go from one to 150, beating Sierra by one quarter.
Answered around 31:51About a trillion dollars a year goes into legal services, done manually and very fragmented, while software spend on legal technology is roughly 40 billion. That is 4% software against 96% services, which is bananas; the software piece should be much bigger and will grow into that service revenue.
Answered around 34:25No fine-tuning and no general intelligence models — that is called a total waste of time and money, and some competitors attempting a general legal intelligence model are wrong. Narrow models for narrow use cases do work, cutting cost and latency: tabular review means 100 documents times 100 prompts, so 10,000 API calls.
Answered around 48:49“Then it took us roughly 20 months to get to the first 100 million in ARR.”
“Today we paid back over $22 million back to the community of talent.”
“So it means there's 4% software, 96% service, which is bananas.”
“we have forward deployed lawyers.”
“Airwallex Revenue Growth and Culture.” https://content.libsyn.com/p/9/3/3/933ee6280af4cbd2/1_Paris_Ch.mp3. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/airwallex-revenue-growth-and-culture.html.