We build precision instruments for people leaders
Paris-built. GDPR-first. Designed for the data-literate HR leaders and COOs who are done with survey averages.
Why we started TeamVyne
Antoine Lefèvre spent four years as an engineering manager at a 300-person SaaS company in Paris. Every quarter, the People team would share engagement scores and ask him why his team's score had dropped. Every quarter, he already knew — but had no way to prove it with data.
The friction was visible if you knew where to look: a daily standup that had grown to 45 minutes, a design review process that fragmented engineering focus, a Product-to-Engineering handoff that reliably dropped context every other sprint. None of this showed up in surveys until it was already costing people.
TeamVyne is the instrument Antoine wished he'd had — not another sentiment tracker, but a precision signal reader that tells people leaders what's actually happening in their team's operational reality.
The team
Antoine Lefèvre
CEO & Co-Founder
Engineering manager for four years at a 300-person B2B SaaS company in Paris. Spent every quarter answering "why did engagement drop?" with observations he couldn't prove with data. That frustration became TeamVyne in early 2025.
Marie Dubois
CTO & Co-Founder
ML engineer with a background in graph analytics and time-series modelling. Designed TeamVyne's friction signal pipeline from the ground up — including the privacy-preserving aggregation architecture that ensures no individual-level data is retained beyond the 14-day processing window.
Lucas Martin
Head of Product
Five years in HR tech product roles before joining TeamVyne — enough time to understand why most people analytics dashboards produce beautiful reports that nobody acts on. Designed the Three Actions output format specifically to make the product's value visible in a single weekly email.
How we work
Signal over sentiment
We trust operational data over stated feelings. Not because feelings don't matter — but because operational data arrives four to six weeks earlier and is harder to game.
Privacy by architecture
We never analyse individual employees. We never read content. Every design decision starts with the question: what's the minimum data needed to compute this signal at team level?
Three actions, not forty slides
Our output format is deliberately constrained. Three numbered changes, ordered by impact, with a predicted metric. If we can't distil it to three actions, we haven't done our job.
Honest about what we don't know
Our models have confidence intervals and we show them. A friction score of 7.3 ± 0.6 is more useful than a crisp 7.3 that implies precision we don't have. We don't inflate predicted outcomes to look more impressive — the cost of a wrong intervention is real, and your trust is worth more to us than a good screenshot.
Work with us on the early access programme
Free during early access. Your feedback shapes the product.
Request Early Access