Over the past year, almost every company has started experimenting with AI in one form or another. Hospitality is no exception.
Hotels are using AI to write content, search information, respond to reviews, prepare reports, handle guest inquiries, and support revenue management.
But there is a big difference between using AI and putting AI to work.
Many hotels have experimented with AI in isolated use cases. Far fewer have embedded it deeply into business processes or built a coherent operating model around it.
That leaves management teams with three practical questions:
Where should AI work? How far should it go? And where should we start?
Answering those questions in the age of AI requires a new map.
1. Don’t ask which jobs AI can replace. Ask which tasks AI should take over.
Whether a task is suitable for AI depends less on the job title and more on the nature of the work.
Review analysis, standard guest inquiries, or daily reporting tend to have several things in common: the rules are relatively clear, the inputs are digital, the work is repetitive, and the output can be checked.
Now compare that with calming an angry guest at the front desk or negotiating a strategic corporate account. AI may be able to participate, but that does not mean it should carry final responsibility.
We use a simple framework we call SPIDER to assess AI suitability across six dimensions: Standardization, Physical execution requirements, Inspectability (verifiability), Digitalization, Emotional risk, and Repetition.
The shift in perspective is important:
Instead of asking, “Can AI replace the front desk?”, ask, “Which tasks within the front desk role can be handed to AI?”
That is a much more useful management question.
2. AI autonomy is not binary
Even within the same task, AI can operate at very different levels of autonomy.
Take hotel pricing.
At one level, AI simply organizes market and demand information. At the next, it recommends a price change. It can then prepare the change for human approval. With more trust and clearer controls, it may execute automatically within predefined price and inventory thresholds.
Eventually, within a well-defined operating boundary, AI could connect forecasting, decision-making, execution, and review into a closed loop.
Borrowing loosely from the language of autonomous driving, I think of AI autonomy in five levels:
L1 Assist → L2 Recommend → L3 Execute with Approval → L4 Authorized Autonomous Execution → L5 Bounded Autonomous Operation.
What changes across these levels is not just the capability of AI. It is the position of the human.
At lower levels, we are human-in-the-loop: without human action, the task does not complete.
At higher levels, we move toward human-on-the-loop or, for routine execution, even human-out-of-the-loop: people define objectives, permissions, exceptions, and oversight rather than touching every transaction.
That is why the hardest part of AI deployment is often not the technology. It is governance.
Who has authority? What are the limits? When must the system escalate? Can every important action be audited?
3. Hospitality needs an AI capability map, not another list of AI products
Most hospitality AI discussions still start from the technology:
AI concierge. AI search. AI revenue management. AI content. AI BI. AI CI (Customer Insights). The product labels and acronyms go on and on.
But hotel operators should probably start somewhere else.
The more useful question is:
What capabilities does a hotel company need to operate — and which of those capabilities are being reshaped by AI?
That led us to build an AI version of a Component Business Model (CBM) for hospitality.
Across the horizontal axis are eight enterprise capability domains: brand and acquisition, sales and accounts, revenue and distribution, development and assets, rooms and F&B, service delivery, supply chain and equipment, and business support.
Vertically, we separate three layers:
Direct — deciding what to do
Control — managing and optimizing
Execute — actually delivering the work
Once AI is mapped onto this structure, a pattern becomes visible.
The closer the work is to standardized, digital execution, the greater the potential for AI autonomy.
The closer it is to major decisions, complex relationships, emotional situations, or physical service delivery, the more human accountability remains essential.
In other words, AI will not transform organizations department by department.
It will transform them capability by capability, task by task.
4. The Bigger Disruption: Rethinking the Hospitality Operating Model
This leads to a more provocative question.
When a hotel opens today, it typically buys a PMS, CRS, RMS, CRM, POS, task management system, BI tools, and more. It then builds departments, roles, approval processes, and handoffs around those systems.
Over time, we have normalized an operating model in which every system has its own interface, every department has its own workflow, and employees become the integration layer between them — moving information through spreadsheets, messages, calls, and manual re-entry.
Humans become the API.
But if we were designing an AI-first hotel company from scratch today, would we make the same choices?
Would we still need as many systems and interfaces?
Would we draw the same process boundaries?
Could we build a thinner technology core, allow AI agents to orchestrate work across systems, and redesign human roles around judgment, relationships, service, and exceptions?
One hypothesis worth testing is that AI’s biggest impact may not be headcount reduction at all.
It may be the removal of system fragmentation, process friction, and coordination work that created many of those tasks in the first place.
If that proves true, the structural advantage will not come from “adding AI” to the traditional hotel operating model.
It may come from designing an AI-native operating model from day one.
So perhaps the question for hospitality leaders is no longer:
“Are we using AI?”
It is:
“Is our AI actually on the job?”
And one question I’m increasingly curious about:
If you were building a hotel company from scratch today, what would you no longer build the old way?