By Tony Loeb, co-founder of 10 Minutes Hotels
When ChatGPT arrived, hotel tech rushed into AI. Everyone wanted to try it, plenty got hooked, and things moved very fast. Looking back, my guest believes a lot of people would have done it differently.
That guest is Kevin King, CEO of Shiji International, who joined me on the podcast. Shiji technology is used by more than 91,000 hotels worldwide. So when he tells me the AI race may have started with the wrong question, it’s worth hearing him out.
Shiji didn’t jump in. It took the time to “look, listen and learn”, a phrase Kevin comes back to again and again in our conversation.
Two things stood out for me: how Shiji tried AI on itself before offering it to customers, and what it should change, in very practical terms, for a front desk team.
A year of internal testing before the first customer
Instead of putting AI straight in front of its hotel clients, Shiji started with its own departments: finance, legal, product development. For Kevin, it comes down to responsibility. Customers rely on their tech provider to be “the custodian of their solutions, their data”. You can’t afford to be reckless with a technology nobody fully understands yet. Or, in his words: “You don’t know what you don’t know.”
That phase lasted about a year. Then Shiji brought in external auditors and its own data protection team (the people who make sure personal data is handled within the law) and asked them to criticize what had been done. Kevin has an image for this mindset. You have to turn to the back page, not stop at the front page “where the good story is”. The back page is where the legal and compliance questions live.
The main risk they found is easy to grasp. When you give data to AI, it may keep that data to learn from. So you need to know exactly what you’re feeding it, and choose models and partners that share the liability.
The caution went a long way. Shiji’s 500-plus developers weren’t allowed to use AI until the end of last year. An AI committee spanning security, compliance, development and product now oversees how it’s used. Kevin sees a clear payoff. Because Shiji had tested everything on itself, it was well ahead of the EU AI Act, the European regulation governing the use of artificial intelligence.
Today’s AI is a fourteen-year-old
I wanted to push Kevin on one point. If ChatGPT gets my chicken recipe wrong, it’s annoying. If AI changes something in the PMS across a hundred hotels and gets it wrong, that’s a real problem.
His answer is a comparison that some people laugh at, he admits. To him, AI today is a 14-year-old child. Fourteen-year-olds are reckless and fearless. They charge ahead, fall over, scratch a leg and get back up. At sixteen, they start having opinions. By eighteen, some “sense and sensibility” comes in. Shiji chose to go through those stages with its own teams.
Kevin doesn’t claim to be infallible. No one can guarantee that any technology will always work exactly as designed, he says. What you can do is put every possible safeguard in place to limit the damage, even if that means not being the fastest to market with a new feature.
A late checkout without the head-down routine
I asked Kevin what changes when a guest asks to check out at 4 p.m.
Kevin travels 300 days a year, so he knows the scene by heart. The guest asks. The answer comes back: “Let me check if that’s available.” The employee looks down and clicks through five different screens to find out who the guest is, which room they’re in, whether they’re a loyalty member, whether they’re entitled to a late checkout. Then they look up and apologize: it’s not possible. Meanwhile, the guest is left wondering whether they’ve done something wrong.
With what Shiji calls microflows (small automated sequences of tasks), the employee simply enters “room 101, 4 p.m. checkout”, or asks the AI in plain language. The system runs the checks and comes back with a green tick. It also notifies housekeeping and prompts the team to extend the guest’s key card.
Until now, staff had to understand the software. Now the software understands what staff are trying to do. Kevin agrees: “Don’t let the technology control you. You control the technology. The complexity is in the back. No one needs to see the complexity.”
The same thinking guided him in Macau, at casino operator Melco. Its leaders gave him one priority: get rid of the long lines snaking through the lobby like an airline check-in. By simplifying arrival and letting guests check in before they even got there, Shiji cut check-in from five to eight minutes down to under one. In a casino, Kevin points out, time in line is money: “They don’t want them in a line, they want them on the floor.”
Understand it before you roll it out
Kevin King’s case comes down to one idea. Get to know AI at home before handing it to anyone else, then use it without complexity so teams can finally have a real conversation with the guest.
In the full podcast episode, Kevin also talks about opening a hotel in Riyadh with staff who had never worked in hospitality, and why he hires on attitude. He explains how casino finance teams cut their balancing work from more than a day to under an hour, the one red line AI must never cross with guest data, and what you’ll no longer see when you walk into a hotel five years from now.
🎧 Watch the full episode here: 91,000 Hotels Use Shiji. Its CEO Says We Got AI Wrong
