Ava AI was generative interviewing at enterprise scale: a voice AI that could run a real, adaptive conversation with a candidate and hand the hiring team a structured read on the other side.
It ran more than 20,000 interviews a month. Completion held at 95%, candidate sentiment at 92%, with Spotify as an anchor client. I was co-founder and CEO of a distributed team of thirty-odd people. It is still the clearest evidence I have that a voice agent can carry a conversation that matters, not just a demo.
It started as a conversation with a recruiter about what an AI recruiter could actually look like. The first round of hiring is a volume problem dressed up as a judgement problem. Thousands of first conversations, and nobody has the hours to hold any of them well.
Making a machine interview worth a candidate's time. Completion and sentiment were the entire game, the moment it felt like talking to a form, people abandoned it. Reaching 95% completion came from obsessing over latency, turn-taking and warmth, not over model benchmarks.
Closed. The AI-recruitment category crowded very quickly, and our differentiator, AI-led interviewing, was being pursued by substantially larger incumbents, Eightfold and LinkedIn among them. Continuing meant an uphill fight with no clear strategic advantage, so we chose to close rather than force a pivot. Several people have suggested a relaunch since; perhaps not the final chapter.