Operating Model · April 2026

How Nilo runs.

35 humans. 6 agents. One shared inbox.
Booking.com runs ~27,000 people to do what we do with a team that fits in one bakery.

The team

35 people. Here's every one of them.

Six PMs (one per agent). Fourteen engineers. Six Field Ops on the road in Lisbon, Porto, Barcelona, Madrid, Valencia, Rome. Three designers who ship. Three on finance/legal/compliance. Two founders. That's it. No marketing team. No content ops. No SEO team. No enterprise implementation org.

Revenue per employee, at Series A trajectory
€2.1M
Target €3M+ at scale. Benchmark: Cursor ~€8M, Midjourney ~€4M. Booking.com: ~€0.8M.
Engineering
14
Foundation-model-fluent. Everyone ships prod code weekly, including founders.
Product (PMs)
6
One PM per agent. They own the evals and the outcome metric.
Field Ops
6
On the road in PT/ES/IT. Onboard a hotel in one morning. Part engineer, part hotelier-whisperer.
Design
3
Design in code. No Figma → ticket relay.
Finance / Legal / Compliance
3
Thin by design. Compliance agent does the filing.
Founders
2
Still in the codebase. Still in the hotels.
The agent stack

Six agents do what four floors of Booking's Amsterdam office do.

Each agent has one PM, one eval suite, one outcome metric. They run continuously. A human steps in only where judgment beats automation.

H
Handshake agent

Runs the pre-arrival flow with each guest — ID, payment, preferences — then hands the hotelier a morning brief. Touches 100% of bookings.

D
Demand agent

Places each hotel inside Claude, Gemini, Perplexity and traveler-side agent answers. Owns agentic-sourced GMV. €11M/quarter today.

F
Fraud agent

Screens every booking for chargeback risk, ID mismatches, stolen-card patterns. 0.07% chargeback rate, 1/18th of industry.

C
Compliance agent

Files local tourism tax, registers guests with SEF (PT), Mossos (ES), Portale Alloggiati (IT). Invoicing in the right local format.

M
Migration agent

Imports the hotel's existing listing, photos, calendar and 3 years of reviews. Rewrites the Nilo listing and re-syncs weekly against booking data.

R
Review agent

Nudges each guest 24h after checkout for a direct Nilo review, turns responses into a property trust score used by the Demand agent.

How we charge

You pay when the booking lands. Not before. Not for impressions.

One event triggers the charge: guest checks in. Everything else is free.

Event
Guest checks in
Confirmed by the PMS. Nilo charges the traveler €9 flat. The hotel pays €0.
No check-in
€0 to Nilo
No-show, cancellation, fraud reversal: we take nothing. We only get paid when the hotel delivers.
Lookup, browse, calendar sync
€0 forever
No listing fee. No featured-property fee. No subscription. No channel-manager fee. No PPC auction.

The hotel never sends us an invoice. The traveler's €9 lands with Nilo at check-in. The hotel keeps every other cent the traveler paid.

Unit economics

The math, unglamorous and specific.

Revenue per completed booking
€9
Flat. Paid by traveler. Same for a €60 hostel or a €400 boutique.
Gross margin
74%
Payment processing, hosting, agent inference costs. Moves to ~82% as inference keeps falling.
Hotelier CAC · payback
4.1mo
Mostly Field Ops salary + a morning of founder time. Paid back by month 4 of the hotel going live.
Hero metric
Bookings that close without a human touching them
92%
End-to-end: discovery → booking → Handshake → check-in. A human touches the other 8% (fraud review, exception handling, exotic compliance).
The absent teams

What we don't have.

The three largest functions at Booking.com are the three we decided not to build. Each one replaced by a specific mechanic, not a framework.

Skipped team
SEO · Affiliate · Performance Marketing
~1,500 people at Booking.com
What we do instead
The Demand agent.

We don't chase Google rank. We don't auction against Expedia for keywords. The Demand agent gets hotels surfaced inside Claude, Gemini and Perplexity answers. Travelers arrive through agentic discovery, not search. Today 38% of Nilo bookings arrive cited by an AI.

Skipped team
Content Ops · Listings QA
~500 people at Booking.com
What we do instead
The listing writes itself.

Hoteliers upload photos and a few sentences of notes. A generative listing pipeline writes and re-writes the page weekly, informed by booking data — which rooms convert, which phrases travelers search, which photos get expanded. No human content team. No content calendar.

Skipped team
Enterprise Implementation · RFP · Chain Sales
Hundreds of people across Booking's chain-hotel org
What we do instead
We only sell to independents.

Nilo sells only to independent hotels with under 30 rooms. No enterprise SOWs, no RFPs, no 6-month rollouts, no procurement committee. Hoteliers self-serve in one morning via the Migration agent. That constraint is what makes 35 people work.

How a decision flows

Tuesday morning, 06:47. A real flow.

A guest flags their flight is delayed 4 hours. Their pre-paid €148 room night at Casa Boavista would be wasted. Here's the full chain — no humans in the loop, except the hotelier, who finds out when she opens the blinds.

06:47
Signal
Clara's inbound flight AA74 is delayed 4h.
Handshake agent pulls flight status via FlightAware at 15min intervals for the 48h before arrival.
06:48
Handshake agent — decides
Texts Clara: "Your check-in is pushed to 19:30. Arrival taxi re-booked. No action needed."
Below the €500 autonomy threshold → auto-executes. Above it, a human would approve first.
06:49
Compliance agent — acts
Updates the SEF registration for Clara to reflect the new arrival time.
Portuguese immigration requires same-day registration. Agent files it; no hotelier work.
07:02
Human touch — Maria opens Nilo
Maria sees one updated line in the morning brief. Clara: 19:30 instead of 15:00. Done.
Maria moves Clara's welcome port wine to a later slot. Total decision time: 4 seconds.
19:34
Outcome
Clara walks in. Port wine ready. Room 4. "You must be Maria?" "You must be Clara."
Zero internal meetings. Zero escalations. Zero tickets.
Humans only arrive when taste matters — the port wine. The rest is mechanism.