Cloudbeds' Josh Graham on why AI finally favors the independent hotel

We asked Josh Graham, Cloudbeds' head of market development, how he describes his job at a party, to someone with no connection to the business. He keeps it to one line: you know when you check into a hotel and the person behind the desk is typing on a computer. That's what he does. Push him a bit further and he says Cloudbeds makes software that helps a hotel run its whole business and drive its revenue. Then he catches himself, because that's an odd thing for a PMS to say, and Cloudbeds doesn't think of itself as a traditional PMS. Josh thinks the PMS is history.

He says the PMS was built to solve a 1990s problem: how do you digitise the front desk. He started in the industry in the late nineties, when plenty of hotels still had a physical tape chart in the back and reservations written in a ledger. The PMS fixed that, and fixed it well. But it was a system of record, built for the things you keep history for, like how you pay your taxes. It stayed there. As SaaS arrived and innovation happened around it, the PMS sat still. You replaced it every three to five years like a worn carpet, spent a pile of money, got a new version. Having your data scattered was an inconvenience first, then an expense. In an age of AI, what matters is unified data under one model a machine can actually read. He says scattered data becomes fatal.

Staging a funeral for the PMS

Josh wanted to stage a funeral for the PMS at HITEC. The idea was to bury it in public and say out loud that the PMS as hotels knew it is dead. For a company most of the floor would still file under PMS, that was a bold move to even propose, and we were a little sorry the marketing team talked him out of it. Underneath the theatre there's a distinction. Cloudbeds started as a unified platform on a single data model, before it even had all the integrations, because that's how the founders built it. A lot of rivals are coming at it the other way, buying companies and integrating them, what he calls an acquisition-and-integration strategy. He did exactly that for 13 years at TravelClick, and he says integrations are messy and never go as cleanly as the due diligence promised. Everyone wants to be a platform now. The difference, he said, is that the others are only arriving at that model lately, while Cloudbeds was built that way from the start.

He doesn't oversell it. Hospitality is 10 rooms on a beach in Thailand and 1,000 rooms in a city with 100,000 square feet of meeting space, so there's no one answer for everyone. If he ever tells you Cloudbeds is right for every hotel, he said, don't believe him. The market backs that up: there are more than 350 PMS systems worldwide, and it stays fractured for real reasons, most of them about localisation. The police report a hotel files in Italy looks nothing like the one in Albania, and the software has to match each local format. Language is the same problem, and Cloudbeds does little business in Japan because it hasn't built the Japanese-language support that market needs. So the question isn't all-in-one versus best-of-breed. It's some of the best. A 75-room hotel in downtown Berlin, dynamic pricing, few repeat guests, plugs a strong RMS into the platform. A 75-room leisure property up a Swiss lake plugs in a strong CRM to drive repeat direct bookings instead. You pick what you need.

Why the quality of an integration is the whole game

That leads to his point on integrations: all integrations are not created equal. It isn't just one-way versus two-way. It's how big the pipe is and how many data points actually move through it. Take a spa: most spa integrations let the spa post a charge to the room, which a front-desk or marketing person doesn't really care about. What you'd want to know is that the guest likes a shiatsu massage from Olga with the lavender oils, so that when Olga has a slot next week, the hotel can message them. That data always existed, sitting in a data lake where it never reached the desk in any useful form. Built in carefully, AI can bring that context up at the moment someone can use it.

What hotels should actually be doing in 2026

His advice to hotels for this year is the opposite of the hype. The winning hotel in 2026 isn't the one with an agentic booking flow. It's the one doing the boring groundwork: taking an inventory of its tech stack, making sure it isn't building data silos, consolidating vendors, and moving its people from knowing what AI is toward being able to use it. He draws a line between AI literacy, knowing what an LLM is and how generative differs from agentic, and AI fluency, actually putting it to work. Ask the HITEC floor what an agent is and he reckons half could answer. Walk into hotel operations and the number drops, which he thinks is fine, because nobody good takes a front-desk job for the love of technology.

He says this is the first technological shift in 30 years that rewards context rather than scale. The internet rewarded the big OTAs and brands. So did the cloud. Mobile rewarded Airbnb for building the best experience. Every one of those favoured size. AI is different, because what it rewards is context, and context doesn't live with the brands or the OTAs, who don't know what actually happens on a property. It lives with the operator. Which is why, he thinks, there's never been a better time to be an independent hotel.

He rejects the opposite view, which we'd heard elsewhere at the show, that hotels already missed the window. He calls it defeatist. Yes, a big brand might have early access to a frontier model that a 75-room property doesn't. But what worked three months ago already doesn't matter, the way it didn't in the early days of search, when the rules changed constantly. He says to stop worrying about the micro question of whether you own this or that tool, and pull back to the macro one: how unified is your data, how unified is your stack, can you consolidate vendors, and of the ones you keep, which actually share their data well. He'd make that last point a deciding factor, even over a longer feature list.

Guessing is not allowed

Cloudbeds' own AI is Signals, which Josh describes as a causal intelligence layer built into the platform. The first thing they switched on was revenue intelligence, which joins two functions that usually sit in different offices. A revenue system has two levers, rate and inventory, but there's a third that moves demand, marketing, and it normally lives in a separate system with a separate person and separate metrics. Revenue intelligence connects them, so instead of dropping a rate by $30 to fill a soft week, a hotel can push a marketing campaign first, then maybe shave 10 or 15 off later if it still needs to. He pointed to Hotel 1550 near San Francisco airport as one of plenty running it.

At their Compass event last month they launched Ask Signals, a conversational layer you talk or type to, with a memory. The difference he claims over similar tools is reach: most see part of the data, like looking through a keyhole at one slice of the room, while Ask Signals sees everything in the platform, the PMS, marketing, channel management. A GM can ask how next weekend looks, or to tell them more about the guest arriving tonight, and act on the answer in the same breath, lowering a rate with a tap.

The discipline underneath it is one we heard from a few vendors this week. Josh calls it probabilistic versus deterministic. A probabilistic model is wonderful for writing an email, where it guesses what you meant and you fix what you don't like. But ask it your occupancy and you're at 75 percent or 72, not whatever number it feels like inventing. Guessing is not allowed there, and if a system like that loses a hotelier's trust, it's done. How far to let it act is a question of the hotel. Where the GM does revenue management for half an hour at night, letting the AI make the calls is probably the smarter, faster option. For a property with a certified revenue manager juggling transient and group business, the individual bookings and the block bookings, it should suggest, not decide.

Will anyone trust it

So does he think hotels will end up trusting the machine. "If I knew, I'd be a much richer man," he said. His answer is that he doesn't know, because the real question is whether people will trust a machine at all. He watches his 16-year-old daughter, who is firmly anti-AI and tells him to stop talking about it, and sees the same in her friends. He won't romanticise it or pretend he can predict how it lands. Maybe it all works and everyone decides it's great. Maybe some of it crashes and people sour on it. He's an optimist about the operator's hand in this, and done overselling the rest.