IMPT operates a hotel platform spanning 8 million hotels and apartments across 195 countries, funds verified carbon offsetting on bookings, runs marketing in dozens of languages — and does it from Clonmel, County Tipperary, with a team small enough to fit around a dinner table. That last fact is the one other founders ask about, so here is the honest anatomy of an AI-first company.
AI does the operations, not the mission
The division of labour is simple to state. Software — increasingly, autonomous agents — does the operations: front-line sales conversations in the customer's language and currency, payment follow-ups, meeting scheduling, content production across thousands of locality pages, monitoring, reporting. Humans do the mission: what we sell, what we refuse to claim, which climate systems we plug into, who we trust. Every failure we've had came from letting that boundary blur — an agent making a judgment call, or a human doing coverage work at 3am that software should have owned.
Scale without head-count is a discipline
The previous generation of travel companies scaled by hiring. The AI-first version scales by specifying: writing down, precisely, what good looks like — the tone of a follow-up, the facts an agent may state, the exact conditions for escalation — and then holding the machine to it with verification. The specification is the company. It is slower to write than a job ad, and it compounds in a way head-count never did. When we fixed how prices are quoted in a buyer's local currency, the fix applied everywhere, instantly, forever.
Verification is the real AI moat
Anyone can deploy a model; the operating advantage is the harness around it. Ours has hard rules: an agent may never claim an unverified action, never state a price from memory, never touch topics reserved for humans. Every send is receipt-checked, every fact tool-fetched. This sounds bureaucratic; it is the opposite — it is what lets us give the machine real responsibility, including revenue-carrying conversations, without lying awake. Trust in AI is not a feeling. It is an audit trail.
Why Clonmel, why Ireland
There's a romantic answer — home — and a practical one that matters more. AI-first companies are location-independent in their operations but location-rich in their judgment. Ireland gives us EU standards to build against (which is why a European regional government's climate-credit system, Catalonia's, became our verification anchor), a time zone that bridges American and Asian markets, and a credibility with both. The Sunday Times wrote about what we're building; the $30M raise is accelerating it. Neither required an office in a capital city — they required the thing to be real.
The founder's job, revised
When software does the work, the founder's job collapses into three verbs: decide (the claims, the standards, the boundaries), specify (so the machine can execute them), and verify (so you know it did). Everything else — the dashboards, the busywork, the performative hustle — the machine either does or exposes as unnecessary. It is the most honest management structure I've worked in: the company is exactly as good as what you wrote down.
Common questions
What does 'AI-first' mean in practice at IMPT?
Autonomous agents and generated systems run the operations — sales conversations, follow-ups, scheduling, content, monitoring — while humans own judgment: claims, standards, escalations and partnerships.
How does a small team run a platform with 8 million properties?
By specifying rather than hiring: precise written standards for what good looks like, executed by AI systems and enforced with verification — receipt-checked actions and tool-fetched facts.
Where is IMPT based?
Clonmel, County Tipperary, Ireland — with EU standards, a bridging time zone, and the Catalan government-supervised climate-credit system as its verification anchor.