Four categories, and what they lack
The status quo: the exceptional maître d'hôtel
He is who we truly measure ourselves against. He makes the move, but he is rare, expensive, impossible to replicate across ten locations, and his know-how is written nowhere. The day he leaves, the venue drops a notch.
Reservations, POS and CRM
Zenchef, SevenRooms, TheFork, Sunday, Toast. The leaders are as American (Toast, SevenRooms) as European (Zenchef, TheFork), already installed in French venues. Very good in their role: they analyze service once it's over. None, here or abroad, steps onto the floor during the rush.
Generic AI in the cloud
Neither local, nor present during service, nor trained on the trade of the floor. A generalist model knows neither the pace of a plating, nor a dish pairing, nor the rhythm of a brigade.
Real-time coaching in call centers
Cresta, Gong. The mechanic is close (whispering to the operator during the exchange), but the stage is the B2B phone: a headset, a screen, a fixed desk. Not a floor, not a brigade, not hands full.
“What if a giant did it tomorrow?”
It's the question every serious investor asks. Here is our answer, plainly.
It's not a feature, it's another company
Real-time in-ear coaching takes four disciplines in one product: embedded AI, sub-second real time, hardware placed on the floor, and twenty years of hospitality field craft. A SaaS vendor holds one. The other three don't get added by an update: it's a whole company to build, not a feature to ship.
The corpus can't be bought
No public dataset contains a room in full service: neither the pace of a plating, nor a dish pairing, nor the word that saves a tip. This corpus can't be scraped, it's lived, one service after another. The day a giant starts, it starts from zero nights of service; we'll have a corpus it will take ~18 months to rebuild.
Included hardware locks in the base
Every box installed removes a location from the market of those who follow, for years. To win it back, a competitor would have to beat a price we've already set at zero. You don't attack an already-installed standard with a better demo: you'd have to beat free at entry.
Sovereignty is decided on day one
An AI designed for the cloud doesn't become local by press release: zero-cloud is decided at the first line of architecture, or never. A cloud player wanting to copy us would have to sabotage its own model: the data it resells, we seal inside the walls.
Environ dix-huit mois : the time it will take an entrant to rebuild the field corpus we'll already have begun.
The real issue is adoption
In meetings, the real question isn't competition. It's adoption: how it's installed, and how teams take it up.
“My teams will refuse”
No individual scoring, no data leaving. The coach advises, it doesn't report. A companion, not a judge.
“This is surveillance”
Data stays within the walls. Video reads the rhythm of tables, not individuals, and erases itself automatically.
“I don't have time to install”
Hardware is included and installation is handled. A dedicated install lead is a line item of the round.
The pilot will measure real adoption: actual earpiece wear, team satisfaction, retention. Those are the numbers we'll bring back, not promises.