AI in PMS and event systems: what does it actually mean?

AI in PMS and event systems: what does it actually mean?

We wrote in our market overview that AI is no longer a differentiator — it's an expectation. Every vendor now says they have it. That makes the question "do you have AI" fairly useless to ask in a demo, since the answer is always yes. The interesting question, which most of us don't quite know how to phrase yet, is: what kind of AI, and what does it actually change in an ordinary operating week?

This post is an attempt to answer that concretely, without either overselling it or dismissing it. Both are tempting, and both are imprecise.

Two quite different things are being called "AI" right now

Most of the confusion comes from the industry using one word for two fairly different things.

The first kind is what we'd call suggestion AI. It reads something (an email, an inquiry, a history) and suggests something back — a draft offer, a recommended rate, an answer to a question. A human looks at the suggestion and decides whether to use it. This is useful, and it's where most of what's sold as "AI features" today actually sits.

The second kind is AI agents. The difference isn't that they're "smarter" — it's that they have access to do something, not just suggest it. An agent can pull occupancy from the system, check a room block, draft a contract, or send a follow-up, because it has tool access into the actual systems, not just into a chat box. It's this second kind that actually changes something structurally, and it's also the one that places entirely different demands on the system it's connected to.

We wrote in our previous post on API access that an AI agent is only as useful as the tools it's allowed to use, and that the Model Context Protocol (MCP) is the open standard that lets an agent call your systems as structured, authenticated tools. It's worth repeating briefly here, because it's the actual glue between the two kinds of AI: without a proper API and a way to expose it to an agent, "AI agent" is just a suggestion with a fancy name — it can never get past the step where a human has to type the result in manually anyway.

Why agents are actually a game changer — not just a buzzword

It's easy to grow skeptical of the term "game changer," since it gets applied to everything from genuine breakthroughs to entirely marginal tweaks. So let's be concrete about why this is actually different this time, specifically for PMS and event operations.

Most of the time lost in this industry isn't lost because someone made a bad decision. It's lost in the handoffs — between an email and a system, between a spreadsheet and a phone call, between what sales knows and what reception knows. A suggestion AI helps with the task itself, but someone still has to carry the information between systems. An agent with real tool access can actually close that gap: it reads an email, creates a lead directly in the system, checks availability, and has a structured draft ready — without a human being the glue between the four steps. The human comes in to approve, not to carry out every step in between themselves.

That shift — from "AI that helps you do the task" to "AI that does the task, and you approve the result" — is the actual change. It's not that the agent is more impressive in a demo. It's that it removes an entire layer of coordination work that used to require a human at every single step.

How to tell real agent functionality from AI glued on top

Given that everyone says they have AI, it's worth having a few concrete questions ready when evaluating a vendor, instead of relying on how convincing the demo was:

Can the AI feature actually perform an action in the system — create, change, send — or does it just suggest something you still have to type in yourself somewhere else? If it's the latter, that's a suggestion feature, not an agent, and that's fine, but it isn't the same promise.

What API sits underneath it? Ask to see it, don't just be told it exists. We've written more on this before, but in short: without a documented, versioned API specification, it's hard to trust that an agent actually has the access the vendor claims.

What happens when the agent gets it wrong? This might be the single most important question. Any vendor can show you the agent working perfectly in a demo. Ask instead: what does it look like when it misunderstands a request, and where in the process is the error caught — before something goes out to a guest, or after?

Is there a human in the loop where it actually matters? Not just as a general assurance in a sales pitch, but concretely: which actions require approval, and which happen automatically? There's a real difference between "suggests a rate, you approve it" and "sends a confirmation to the guest automatically," and you should know exactly where that line sits for every feature you're considering turning on.

What determines whether you actually get anything out of it — internally, not at the vendor

Here's something that easily gets lost in the AI conversation, but which we think matters just as much as the technology itself: an agent is only as good as the data and processes it gets to work with. We wrote about exactly this in our post on implementation — clean data, clear routines, and an organization that actually knows how things are supposed to be done are a precondition for automation to work, not something that falls into place on its own just because the system is smart.

Concretely, that means: if your pricing logic today lives in the head of one experienced salesperson and has never been written down consistently, a revenue agent can't guess that logic any better than you've managed to articulate it yourselves. If your room codes don't match the physical building after a renovation three years ago, the agent inherits exactly the same confusion the rest of the system has lived with since then. AI doesn't fix a mess. It makes the mess faster.

It's also worth thinking through, internally, how much trust you're actually comfortable handing over first. Most people we talk to start with low-risk tasks — let an agent draft offers, but not send them, for a few months — before gradually letting it take over more of the steps in between itself. That's a sensible order to go in, and there's no shame in moving slowly here, even if a salesperson tries to tell you your competitors have already turned everything on.

So, concretely

Everyone will say they have AI. That makes the question less interesting than it used to be, and shifts the conversation to where it actually belongs: is this an agent with real tool access, or a suggestion with a fancy name? What happens when it gets it wrong? And — perhaps the thing you can actually do something about right now, regardless of which vendor you end up with — are your data and routines actually ready for something to be automated on top of them?