Every PMS pitch this year includes the word AI, and the word now attached to it is native. In the first two weeks of October alone, Shiji published a framework for telling AI-native systems apart, Mews launched AI guest messaging and no-code automations, and a Singapore start-up called PHAL launched what it calls the world's first Total Hotel agentic AI solution.
For hotel tech vendors, the important change is not the launches. It is that buyers now have a test. Once a claim can be tested, it stops being a message and becomes something you have to prove in the demo. Here is the comparison, using only what vendors and studies published.
A working definition
An AI-enhanced product uses AI to do individual tasks faster. An AI-native product makes AI part of the layer through which work gets done: it reads live operational context, acts within permissions, and records what it changed.
That definition follows the framework in Shiji's analysis, Hotel Software Is Learning to Understand Intent (7 October). Shiji describes three levels and admits they overlap, which is exactly why it says counting AI features is a poor measure of how much a system has changed.
Level | What the AI does | Example in a hotel |
|---|---|---|
AI-enhanced | Uses AI for individual functions, such as summarisation | Summarises a guest profile or a shift handover |
AI-assisted | Understands operational context and recommends what an employee should do next | Prepares recommended room assignments for staff to review |
AI-native | Makes AI part of the interaction and orchestration layer through which work is carried out | Prepares or carries out a late checkout within permissions, and logs it |
The five-step spectrum buyers will use
The most useful line in the Shiji piece for sales teams is a progression: AI answers, recommends, prepares, executes, then potentially monitors and adjusts. Each step needs more access to hotel systems and stronger controls.
Step | What it means | What the buyer needs to see |
|---|---|---|
1. Answers | Explains a policy or retrieves information | Answers grounded in the hotel's own data, not generic text |
2. Recommends | Suggests the next action from live context | The data behind the suggestion, visible to staff |
3. Prepares | Builds the change for a person to approve | A clear approval step and the ability to edit |
4. Executes | Carries out the change itself | Permission checks, charge rules and a full audit trail |
5. Monitors and adjusts | Re-plans as conditions change through the day | Escalation rules for conflicting or unusual cases |
Shiji's late-checkout example shows why the jump from step 3 to step 4 is the hard one. Automatic execution means checking whether the action is permitted, whether a charge applies, and whether the extension affects another reservation or the housekeeping schedule. As the analysis puts it, good AI architecture is partly about knowing when not to act.
Where October's launches sit, by their own description
These placements use only the vendors' own published words. They are not a judgment of the products, which buyers should test themselves.
Launch | What the vendor says it does | Closest step, as described |
|---|---|---|
Mews Guest Messaging and Automations (1 October) | AI trained on the property's own data answers routine questions; staff set rules once in a no-code workflow engine | Answers, plus rule-based execution set by staff |
PHAL OS (13 October) | A single agentic AI reading performance, profit, preference and people across the asset's life; agents read and recommend, people decide, nothing executes on its own | Recommends, by design |
Shiji framework (7 October) | Describes the path from answering to executing and argues for gradual progression | A test, not a product claim |
The details are worth knowing because buyers will quote them back to you. Mews says one in three guest messages goes unanswered because they land across six or seven inboxes, that guests who do get a reply wait eight hours on average, that Mews AI answers 80% of routine questions instantly, and that 47,000 automations ran during the beta. At selected Strawberry properties, up to 60% of upgrades for Gold and Platinum guests were automated. PHAL says a hotel department head loses close to two hours a day to spreadsheets, which by its own calculation is some 200,000 hours a year across four department heads in a 100-hotel portfolio, and that a third Top 10 global hotel group has selected it for a proof of concept.
The most specific numbers in both announcements are about time returned to staff, and the most careful wording is about control. That is a positioning choice, and it matches what buyers worry about.
What buyers are worried about
The h2c AI Opportunity Study 2026, based on 113 hotel chains, found that 91% of hotel chains already use AI but only 28% have a company-wide AI strategy. Only 13% report a measurable return on investment. On autonomous decisions in guest conversations, 56% worry about unclear liability for errors, 54% about privacy and security, and 52% about accuracy and bias. I covered the full study in six myths about selling AI to hotel chains.
Put that next to the spectrum and the sales problem is clear: most buyers cannot yet prove the value of step 1 and worry about who carries the risk at step 4. A pitch that jumps straight to autonomy meets a buyer not ready to sign for it.
What buyers will test, claim by claim
Shiji closes with four questions: can the AI work with current hotel context, can it take part in workflows while respecting permissions, can employees inspect and override its actions, and is there an audit trail when it changes a record. They are easy to ask and hard to fake.
Buyer question | What an AI-enhanced product can honestly say | What an AI-native claim has to show |
|---|---|---|
Can it see current operational context? | It works on the data it is given, for a defined task | It reads live room status, arrivals and permissions on demand |
Can it take part in workflows? | It speeds up a step a person still performs | It prepares or executes actions inside the workflow, within user permissions |
Can staff inspect and override it? | Staff make every change, so yes by default | Clear points where staff approve, override or take over |
Is there an audit trail? | Normal user logs apply | Every AI change is logged: what changed, why, and on whose authority |
What happens with bad data? | Staff review the output | Rules for duplicates and gaps, because it may act on them |
The data question is the hidden deal risk
An experienced employee may recognize an incorrect guest profile and disregard it. An automated system could treat the same information as legitimate operational context.
Shiji's conclusion is that AI readiness begins with data readiness, and that a PMS migration is the moment to decide which records to cleanse, which to move and which to leave behind. For vendors, that turns data cleansing from a services line item into part of the AI pitch. If your migration plan says nothing about data quality, an informed buyer will ask, which is why a strong migration plan already closes deals.
How to position, depending on what you actually have
If your product is... | Lead with | Avoid |
|---|---|---|
At step 4 or 5 | A live demo of one real task end to end, with the permission check and the audit log on screen | Feature counts, which the framework teaches buyers to discount |
At step 1 to 3 | The specific job, the time it saves, and the approval step that keeps staff in control | Borrowing the word native; one failed test in a demo costs the deal |
Not a PMS, but plugged into one | Which PMS data you read and write, and what your AI does with it safely | Implying your AI acts on PMS data you cannot reach |
The third row covers most of the market. Revenue management, messaging, upselling and housekeeping tools all depend on PMS data. Shiji notes that a single shared database is not required, only controlled, interoperable interfaces. That makes your integrations part of the AI story, and integrations already sell hotel software more than features do.
The claims are spreading beyond hotels. A VRM Intel column in early October argued that AI-native platforms will move the vacation rental PMS from a system of record to a system of action.
What to do this quarter
Place your product honestly on the five-step spectrum and make sure sales, marketing and the website use the same step
Build the demo around one real hotel task, start to finish, with the approval step visible
Prepare answers to the five buyer questions above before they are asked
Lead with time returned to staff and with control, the two things October's launches chose to quantify and stress
Add a named data quality step to onboarding and migration, and put it in the proposal
If your AI message gets interest but deals stall at the demo or the technical review, the gap is usually in positioning and proof rather than the product. A free revenue leak audit shows where in your pipeline those deals are being lost.
Frequently Asked Questions
What is the difference between an AI-native and an AI-enhanced PMS?
How can a hotel test whether a PMS is really AI-native?
Is AI-enhanced software worse than AI-native software?
Why does data quality matter so much for AI in a PMS?
Should my hotel tech company call its product AI-native?
Get the latest strategies, trends, and expert advice to help your hotel tech business stay competitive and grow.
Where is your revenue actually leaking?
Twelve questions, about twenty minutes, on your own funnel. The same audit I run on day one of an engagement. No call required.
Get the free leak audit
