The companies spending billions to build AI can't tell you if it's working, and the CFO of a $150 billion travel giant just said the quiet part out loud.
The Summary
- Booking Holdings CFO Ewout Steenbergen told Fortune's AIQ Summit that even AI hyperscalers don't know their ROI yet, exposing the industry's metrics problem
- AI is cutting Booking's costs, but LLM-driven bookings account for less than 1% of room nights, a stark gap between efficiency and revenue
- Steenbergen uses an "AI coach" personally, signaling where executives see clear value: productivity tools, not paradigm shifts
The Signal
Booking Holdings CFO Ewout Steenbergen just gave the AI industry the kind of honesty it desperately needs. At Fortune's AIQ Summit, he admitted what everyone suspects but few will say: the hyperscalers building the infrastructure don't know if their massive AI investments are paying off. If the companies selling the shovels can't measure the gold, what does that mean for everyone else digging?
Booking Holdings operates at serious scale. The parent company of Booking.com, Priceline, and Kayak books over a billion room nights annually. When their CFO says AI is cutting costs, that matters. Cost reduction is real, measurable, and happening now across customer service, operations, and internal workflows.
"LLMs still only send the company less than 1% of room nights."
But here's the gap: LLM-driven bookings account for less than 1% of their room nights. Less than one percent. AI is making Booking more efficient, but it's not fundamentally changing how people book travel. The tools are assistants, not replacements. They're margin improvers, not revenue drivers.
This tracks with what's happening across enterprise AI adoption:
- Companies deploy AI for internal efficiency first
- Customer-facing revenue impact lags by years
- The hype focuses on the future, the spend focuses on the present
Steenbergen's own behavior is telling. He mentioned using an "AI coach" personally. Not for booking travel or running financial models, but for productivity. The use case that works is the one where AI helps a human do their job better, faster, cheaper. Not the one where AI does the job.
The Implication
The ROI crisis is real and it's coming for every AI budget. If hyperscalers can't prove returns on their infrastructure investments, and if a company like Booking sees cost savings but sub-1% revenue impact from LLMs, we're heading for a reckoning. Not a crash, but a recalibration. The companies that survive will be the ones solving actual problems, not chasing demo magic.
For people building in this space, the signal is clear: build the AI coach, not the AI employee. Build the tool that makes the existing workflow 10x faster, not the one that promises to replace the workflow entirely. The market for efficiency is infinite and measurable. The market for replacement is theoretical and, so far, mostly imaginary.