Shiji's Natalie Kimball: hotels can't outspend OTAs on AI, but they can still win on experience

We sat down with Natalie Kimball, vice president of strategic account management for Horizon Distribution and Iceportal Content at Shiji, at the company's booth, where they were handing out a printed map of the hotel-tech landscape: the hotel in the centre, and every company that touches its distribution arranged in rings around it, from PMS and booking tools out to AI discovery platforms, social channels, and map-based search. It isn't really about where Shiji sits on it, she told us. It's a picture of the whole industry, and they reach out to every partner on the chart to let them know before it goes out.

We actually tried to keep AI out of it to begin with, and asked whether we could talk about anything else first. Natalie went straight at it anyway. She had strong opinions and no hesitation about sharing them, which made for one of the livelier conversations of the show, and she opened with what most vendors won't say: hotels are losing the same distribution fight they lost to the OTAs 25 years ago, and AI is speeding it up. Back then everyone panicked, handed too much ground to the OTAs, and ended up at 25 or 30 percent margins because they couldn't compete. Her example was a famous landmark hotel that no longer really controls its own name in search, because Booking and Expedia outbid it. She kept coming back to how strange that is, a hotel that can't own its own name.

Why the OTAs start ahead

She thinks AI widens that gap, and quickly. The OTAs can answer a request on a single page, where an individual hotel struggles to. Her point was that the OTAs have captured the specific preferences that decide a booking, the small personal details a traveler actually cares about, while the individual hotel hasn't. So when a guest is matched to the right room, it's the OTA that can do it, because it knows what the traveler wants and the property doesn't.

The reason is the same one that decided SEO. The OTAs spend billions a year on Google, and the same structured data that won them ordinary search is what wins them the AI version of it. They've spent years organising their data so a machine can read it cleanly, and that's exactly what an LLM uses when it answers, so the OTAs start in front and stay there. The question is what happens when someone shops for a hotel inside ChatGPT. Natalie's answer was that the big brands will be fine, the LLMs will learn to speak their language and the MCPs will process them, but the boutiques, the midscale, the roadside place, the property on an Austrian mountainside, they won't show up at all.

What it does to commissions

We raised the question that follows, whether total distribution cost goes up or down, and Natalie worked it out as we talked. Margins probably go up. She guesses the OTAs start splitting their rates by channel, charging more for an agentic booking than for a normal mobile or desktop one, the same way packaged deals already cost more. They did it with packages, hotel plus air, hotel plus car, so they'll do it with AI. They're smart, that's exactly what they do. And hotels will have to get used to paying a lot more commission, which is not something they're ready for.

The narrow window last year

There was a moment, and it passed. Last year's HITEC was the first time hotels could have looked at all this, decided where they wanted to be, and moved on it. There was a window of maybe a couple of months, and if even one of them had executed quickly, the industry wouldn't be in the position it's in now. Nobody did, and the window closed.

She wishes they'd treat software the way they treat the building. Hotels run R&D on the physical product, the door locks that jam, the TV remotes that keep dying, the wear and tear they know how to fix. They just don't do it on the software and solutioning side. Everyone saw AI coming, and it would have been something for a Marriott or an IHG to stand up and say, we're going to try this, and it's okay to try and fail. The problem is everyone wants to get it right out of the gate, so nobody moves. We put it to her that the brands had a chance to take a stand against the OTAs and do it better, and her answer was measured: they'd tried before, more than once, without it changing much.

You can't AI your way into content you don't have

On content, she has a line: "You can't AI your way into correct content." The OTAs are good at collecting the specific details that actually decide a booking. The hotel has that information too, of course. The problem is it never makes it onto the hotel's own website, because hotels don't realise how much it now matters to spell out every specific detail in their own content. And that's the whole game, because the AI answers from whatever content it can find. If a detail isn't stated on the hotel's own site, but it is described on Booking or Expedia, then the OTA is the one that shows up in the answer. The hotel loses the booking by omission.

She took it to a more specific example. Picture a guest who takes one trip a year and wants a rooftop garden with a bee pollinator and honey she can buy made from those bees. No keyword trick gets a hotel into that answer. Either the detail is in the hotel's own content, truthfully, or it isn't.

She'd fill that gap with reviews, which she thinks could become one of the most valuable content sources in AI-driven discovery. We pushed back on that, since reviews can be bought, both the good ones and the bad. She had an answer: at volume, the noise cancels out. Five hundred reviews tell you nothing, but 50,000 that keep landing around 4.7, and keep mentioning the same service or the same breakfast, are something you can trust. She thinks user-generated content gets more powerful in an AI world, not less, because it's the real guest experience turned into bookable data.

TikTok ate the inspiration phase

Another shift is where discovery actually happens now. The industry invested in 360 tours, and far fewer people shop that way now. People want a TikTok video. They see it, they book it, they're done, already planning the next one. The old inspiration phase from 20 years ago is gone. She pointed at meeting planners building a TikTok video and sending it round rather than working through a floor plan, and at how fast people consume it, half a second of footage and they know whether they want more. The next generation doesn't book on TikTok yet, but it's where they go to find the place first, the restaurant, the trip. The attention span for anything slower is gone.

What about Shiji?

Up to this point Natalie had focused more on the broader industry challenge, so we asked what Shiji is actually building for all this. She pointed to the need to connect a hotel's structured content, the visual and descriptive detail that makes it distinctive, with its live availability, rates and inventory, so that what makes a hotel worth choosing travels together with what's actually bookable. Through Horizon Distribution and Iceportal Content, Shiji works to bring those two sides into one place, and she said the last few years have gone into pulling that scattered detail into structured form so a hotel has a fighting chance in the agentic space.

What matters most is that the data has to be correct, so Shiji checks the wider ecosystem against itself. If Booking has just posted a review saying the pool is closed, and the CRS still shows it open, and Expedia shows it closed, the system can weigh those against each other and attach a confidence score: we think the pool is actually closed right now. The work underneath is unglamorous, pulling one hotel's single version of the truth out of seven different sources, the POS for restaurant information, the PMS for room descriptions, the GDS for the address, one system for taxes and shuttles, a flat file for room codes. The bees on the roof, she admitted, are usually stuck somewhere like Virtuoso. Shiji's job is to see all of it and build something coherent out of it.

What she'd tell a hotel to do

Despite the challenges, Natalie's advice to a hotel is calm. Stop trying to solve what you can't solve. You can't change your location and you can't outspend the OTAs, so do the best possible job on the thing you control: a great product, serviced well, available and open. That's where she lands, she's a hotelier at heart, and the thing she keeps defending is the part of all this no model can supply, the connection that made someone want to host a stranger in the first place.