Building a $4B AI video company before the AI hype cycle

Executive overview

Most AI companies chase the hype cycle. Synthesia spent three years before ChatGPT building video AI that barely worked, got rejected by 100 investors, and pivoted twice before finding product-market fit.

The breakthrough came from targeting people who had never made video — not those already making it. Users compared AI avatars to text documents, not real cameras. That comparison changed everything.

The right customer comparison defines your quality bar and your product.

From gimmick to painkiller

• First product — AI dubbing for Hollywood — worked only in perfect conditions and took 10 days per 30-second clip
• Quality threshold was impossibly high: film professionals accepted nothing but the best
• It was a vitamin: interesting, but nobody would scream if it disappeared
• Pivoted to avatar-based corporate video for people who had never made a video before
• Corporate video is a far narrower problem than "any Hollywood film" — that constraint enabled a buildable MVP

Finding real users inside viral noise

• Avatar MVP went viral; most users made a few videos and never returned
• A small cohort kept coming back — those users were comparing AI video to text documents, not to camera footage
• Against that comparison, quality bar and feature requests looked completely different
• Acid test: would customers scream if the product shut down tomorrow?
• By early 2021, a core group passed that test — that was the signal to commit

Combining PLG and enterprise sales

• Focused on enterprise early: highest contract value and clearest ROI
• Growth channel was still bottom-up: let users self-qualify before touching sales
• PLG and top-down conflict on product roadmap, resource allocation, and the type of people you hire
• Best outcomes come from combining PLG's "vortex" acquisition with a world-class enterprise sales motion
• The cultural clash between self-service and enterprise teams is real and requires deliberate management

Hiring underdogs over brand-name recruits

• Hiring from Google or Meta often fails: mismatched salary expectations and startup tolerance
• Better signal: engineers running strong open-source projects, people with hustle who haven't had the "easy" career
• A tight group of hungry underdogs consistently outperforms brand-name hires in the early stage
• Give people room to be shaped — that's the only real option before Series A

Conviction through doubt

• Never doubted the technology would be valuable; did doubt timing and execution
• Being optimistic about early-stage technology is itself an arbitrage — most people default to imagining failure
• Found the one investor (Mark Cuban, cold email) who already shared the vision — didn't need to be convinced
• Finding believers is faster than converting sceptics, especially outside the current hype cycle
• Don't wait to start: four years at McKinsey won't make you a better founder; starting will