Booking rebuilt service and engineering around AI — cost per booking fell (interview)

Speaking at Fortune's AIQ Summit on 1 October 2026 and in an interview with Fortune, Booking Holdings CFO Ewout Steenbergen said the company's first two years with AI were about experimenting, and the next phase means rebuilding processes 'from a white sheet'. Customer service went first.

The results he described are internal and self-reported: bookings are up at a high-single-digit rate while customer service costs are slightly down, so cost per booking has fallen 'by a lot', with customer satisfaction rising at the same time.

In engineering, Booking's roughly 9,000 engineers are getting about 30% more code into production, counting only merge requests that pass testing and quality control. To keep AI spending in check, Booking uses what he calls 'effective model cost routing' — basic and open-source models handle simple tasks, expensive models take complex ones — and engineering groups are measured on total IT cost per merge request, combining human and AI-token costs.

The customer-facing side is still small, he cautioned: LLM referrals were under 1% of room nights as of the Q2 call, and Booking spends $8-9 billion a year on paid channels bringing in about a third of its customers, while its own AI tools show modest gains — faster bookings, slightly higher conversion, fewer cancellations. Q2 numbers he cited: Connected Trip transactions grew low-double-digit at Booking.com, merchant bookings reached about 73% of gross bookings, and Genius Level 2 and 3 members made up more than 30% of active customers.

30% more code into production across 9,000 engineers; service cost per booking down 'a lot' while bookings rose high-single-digit (self-reported)

Customer service went first because it was the clearest end-to-end redesign, and it produced measurable cost and satisfaction gains.

Engineering throughput is counted only on merge requests that pass testing and quality control, so the 30% gain is quality-gated rather than raw output.

Cost routing keeps token spending from outrunning value: teams are measured on total IT cost per merge request, human and AI combined.

Customer-side AI remains small — LLM referrals under 1% of room nights — so the internal process gains are what justify the spending today.

Experimentation is the cheap phase; the returns come when you rebuild a process end to end — and only if you meter the tokens like any other unit cost.

Steenbergen told investors AI will make travel planning easier and marketing cheaper, while insisting the hotel, airline and rental car stay the same; the bigger opportunity is turning a transactional platform into a higher-frequency one through proactive trip management around disruptions. He also disclosed he personally runs two AI agents — a 'strategic thought partner' for board decks and a 'critical equity research agent' for earnings-call prep — and talks about his 'AI coach' openly inside the company. All figures are the CFO's own, from the interview and Booking's Q2 2026 call.