Canary Technologies built its software around the guest, not the property

We asked SJ Sawhney, president and co-founder of Canary Technologies, to explain the company to someone outside the industry, and his answer was an argument. The industry has always treated the hotel as its primary asset, and Canary's whole idea is that this is wrong: the guest is. He walked through the three-letter acronyms to make the point. A PMS is property-focused at its core. An RMS is built around revenue. A CRS is about central reservations. All useful, all necessary, but Canary starts with the guest, and everything it builds comes back to the traveler rather than the room, the building, or the algorithm.

Canary now touches the whole guest journey. It started at check-in, on a kiosk or a phone, and expanded outward into in-stay things like F&B ordering and booking a spa, then earlier into pre-stay, joining the loyalty programme or upgrading a room, and SJ wouldn't be surprised to see it grow into post-stay surveys and reviews, since that's simply the next stretch of the same guest-first logic. He admits this makes the company hard to summarise: mobile check-in isn't really the one core product anymore, because messaging and the rest have grown just as big.

Mobile check-in, and the menu before you land

On how mobile check-in has evolved, SJ's first point: a company can't move faster than the market. Mobile check-in only took off after COVID, through the early experimenters in 2021-22, the early adopters in 2022-23, and into the mainstream in roughly the last 12 months. Canary's approach is never to sit in a basement inventing the next feature, but to watch its customers mature and move with them. One example is acquiring OpenKey earlier this year, the best mobile-key company in the world by his account, timed to when Apple Wallet and Google Wallet had matured and the price point made sense.

He keeps returning to what pre-stay information does for a guest. Knowing three or four days out that the hotel has a restaurant, seeing the menu, screen-shotting it to your partner, knowing there's a kid’s menu. All of that quietly takes away worry that never showed up in a bad review, but shaped the stay anyway. Ten years ago a couple wouldn't have told you they wished they'd seen the menu in advance. Now they have it, and the question of where (and what) the kids will eat is just gone. That, he says, is the "maniacal focus on the guest experience" Canary keeps coming back to: working out what guests are longing for even when they can't put it into words.

Not replacing the front desk

This isn't about replacing staff, SJ said; it's about what he calls "guest optionality." Canary doesn't dictate that you must use a kiosk. Check in on your phone and go straight to your room with a mobile key, use the kiosk, or talk to a human if that's what you want. Its best customers aren't trying to fire their front-desk agents. They're supporting them, so the agent can have an actual conversation, ask where you're from, talk about what's nearby, instead of spending the whole interaction asking for an ID and a card, a transaction that tells the agent nothing about who the guest is or why they're travelling.

A colleague put a number on it: hotels using these touchpoints across the stay see guest satisfaction scores rise 5 to 7 percent, the metric they actually care about. The ID upload is another illustration of the same guest-first thinking. SJ's question is why a loyal guest should hand over a card and passport at every single Wyndham, or every Ace Hotel, when the brand already knows them. That problem only becomes obvious, he says, if you start from the guest rather than the property management system.

Why Canary was ready early

Canary moved early on AI for an uncommon reason: one of its major early investors is also a founding investor in OpenAI. What that gave Canary mattered less as early access to models than as a mindset. AI was on the company's radar well before ChatGPT-3 made LLMs mainstream. He says a company's DNA is the average of the five companies it spends time with, the way a person is the average of their five closest friends, and in the early days of AI, OpenAI was one of Canary's five. The starting point was messaging, a high-communication product where it was natural to ask how much an LLM could do.

He keeps coming back to the idea that you can't cheat the experience. Canary has spent two or three years testing to find out what works best, and that hard-won judgment is what now lets it size up an agentic opportunity and know how to approach it. It's the same instinct, he says, that lets a decades-long HITEC veteran glance at a booth and know exactly where to file the company. He compared it to the early iPhone: the App Store winners, the Angry Birds and Zyngas, weren't an accident, they had proximity to the innovation. Canary's people came from Google, OpenAI, Microsoft AI Labs, and that proximity comes through lunches, not blog posts.

Analyze, assist, do

Here's how Canary thinks about deploying any AI product: analyze, then assist, then do. They don't walk in on day one and take over a hotel's conversations. The AI first analyzes what humans are already doing, then graduates to assisting, telling front-desk or call-centre agents which words to avoid or to ask about the loyalty programme upfront, and only then does it actually do the task. Internally they call it AAD, and he stresses that hotels get a lot of value in the analyze stage alone, before anything is automated.

His ROI case is the voice AI product. Thirty percent of calls to a hotel go unanswered, and a large share of those are people calling to book a room direct, the exact direct business hotels say they want. The AI agent picks up every time. Even for guests not ready to book with an agent, the hotel at least learns there were three booking calls this morning it can ring back, calls that previously just vanished. The second example is the new sales agent: the biggest reason hotels lose meetings and events business is simply not responding to inquiries in time. Canary's sales agent responds to all of them, and analyzes the inquiries a hotel didn't win, so a property where half the business is meetings can finally answer why February was slow instead of guessing. The AI is unemotional, so it doesn't over-index on the one deal a human happens to remember.

The native-app bet

SJ had an analogy for why Canary builds vertical. He lived through the mobile era, when the debate was HTML5 mobile web versus native iOS apps, easy and universal on one side, harder but truer to the platform on the other. He thinks the same split is happening in AI now. The horizontal players just using OpenAI or a general model for everything, are the HTML5 bet: easier, faster to stand up. A vertical, hospitality-specific solution is like building the native app, because there are nuances to hospitality you only learn by building for it, the way there were iPhone nuances you only learned by building on the iPhone.

He flags his own bias here. He's in the vertical camp and admits he's protecting his territory, that a fellow aficionado in the horizontal camp could sound convincing. But he's convinced real industry solutions get built vertically, and he'll let time settle it the way it settled mobile, where in 2010 people were having the exact same argument.

What he won't put AI on, and what's coming next

Asked if there's anything Canary is unlikely to use AI for, SJ's answer was digital tipping, one of their most popular products, because there's simply no clear benefit to  adding it. It fits his line that they don't just bolt the word AI onto everything, especially in a hall where, by his count, 90 percent of booths have "agent" or "AI" on them, exactly the noise that makes it hard for customers to sort real from hype.

He closed on this: there are two revolutions, not one. There's the AI revolution, now climbing the adoption curve, and behind it an agentic revolution that's roughly where AI was two years ago. And he says nobody, including Canary, fully understands the AI-versus-agentic distinction yet. They have strong hypotheses and they experiment with them, but he thinks anyone claiming to know exactly what this is, this early, is overstating it. He'd rather say plainly that it's premature than pretend otherwise.