Operating Model
42 humans, 6 agents, one shared inbox. No category managers. No onboarding teams. No SEO floor. The agents run the operation — the humans run the agents.
Headcount · 42 people
Booking.com: ~27,000 employees. Same product category. 643× the headcount.
Unit economics snapshot
The agent stack
Fetches live supply across all sources in real-time. Every query hits supply directly — no cache, no stale rates. Returns 12,000+ options in under one second.
Executes reservations end-to-end on behalf of any platform. Handles payment, confirmation, and PNR generation. Disruption watch activates on every confirmed booking.
Creates and updates persistent traveler context across all integrations. Seat preferences, dietary needs, loyalty numbers — held once, available everywhere. Cross-platform, never reset.
Handles all CRUD on existing reservations — changes, cancellations, upgrades, seat swaps. Every operation reversible. Audit log on every state change.
Assembles flights, hotels, rail, and ground transport into a single coherent trip thread. Checks connection times, layover viability, and preference alignment before returning.
Proactively monitors for cancellations and overbookings. Alerts and acts before the platform knows there's a problem. Rebooks within approved parameters without human touch.
How we charge
€47 per completed trip. €0 on failures. No commission on inventory lookup. No monthly seat licences. The pricing model is the product promise — if the booking doesn't complete, we don't get paid.
This is not a philosophy — it's the mechanic. Every invoice line maps to a delivered outcome. No retainer. No per-query fee. No gotchas.
Unit economics
87% of trips close on their own — agent executes end to end, no human in the loop. The 13% that escalate are the high-stakes cases: spend over threshold, multi-segment reroutes, VIP travellers. A human reviews within 3 minutes.
What we don't have
Every function below runs at Booking.com with a dedicated team of 20–200 people. At Raile, the same function runs through a mechanic instead.
Decision flow
Real scenario. Amsterdam → Tokyo. CDG connection broken by a 4-hour delay.