Predicting AI is futile, preparing isn’t

There is a lot of thinking going into predicting what AI will look like in our industry. I think that is mostly futile. Ten years ago, very few people would have accurately described today’s AI, and there is little reason to believe we’ll do better predicting 2036. But preparing for that future is actually much easier than predicting it.

Whatever future AI looks like, it will need two things: inputs and outputs. It needs to understand what is happening, which means data. And if it is going to be genuinely useful rather than just an impressive advisor, it needs to be able to do things.

For hotels, the input side (to AI) is relatively straightforward: accessible, structured, reliable, sufficiently real-time data. The output side (from AI) means APIs that allow another system to modify inventory, update a reservation, create a task, communicate with a guest, move a room, trigger a workflow, and eventually do things we haven’t thought of yet. We have been talking about accessible data and APIs for years. AI just makes the reason for having them much more obvious. [Some have said that signing up for my newsletter is obvious as well.]

There is a third piece too: control.

A hallucinating chatbot is annoying. An AI with permission to change reservations, rates or payments can be expensive. Once AI moves from “tell me what I should do” to “do it for me,” permissions, security, auditability, reversibility and guardrails become dramatically more important.

This is why I think the PMS becomes more important in an AI world, but perhaps not for the reason PMS companies would like. The PMS doesn’t necessarily need to contain the intelligence. But it sits close to the operational heart of the hotel. Future intelligence needs a reliable way to understand that environment and safely act upon it.

So, instead of asking, “Which PMS has the best AI?” perhaps hotels should ask, “Which PMS gives me the greatest freedom to use whatever turns out to be the best AI?”

That means APIs and support may ultimately matter more than today’s collection of AI features.

And support is going to matter more, not less. This is slightly counterintuitive. We tend to assume better technology means less human support. But experimentation means things will break. Integrations will fail. Vendors get acquired. APIs change.

Hotels are strangely allergic to technology failure, yet great hoteliers are incredibly good at operational failure (probably the best). Something goes wrong in hotels every day. The skill is fixing impossible problems while making everything appear smooth to the guest.

Perhaps we need to bring more of that mentality into technology.

The objective shouldn’t be to build a stack where nothing fails. If you are seriously experimenting with rapidly developing technology, that probably means you aren’t experimenting enough. The objective is to build one where failures are contained, recoverable and informative.

We cannot predict the winning AI. But we can build a hotel tech stack that is ready to use it when it arrives.

And that may be the much safer bet.