Lead Hospitality

AI Knows a Lot About Hotels—but Almost Nothing About Yours

THE IDEA

When several hotels use the same AI, the difference is not access to ChatGPT but the context each property can provide. This article explores why HotelGEX turns data, business rules and operational memory into a capability to make better decisions, increase revenue and deliver more consistent guest service.

Imagine two hotels opening ChatGPT at the same time and writing almost identical requests. Both want to prepare the morning report, review the day’s arrivals, draft a guest message and identify commercial opportunities. They use the same model, have access to similar capabilities and will probably receive a well-written response within seconds. From the outside, it would appear that both properties have adopted the same capability. Yet one gets a correct but generic text, while the other receives an analysis that recognises its priorities, interprets its constraints and proposes actions that can actually be implemented.

The difference does not lie in the name of the tool or in the elegance of the prompt. It lies in everything the artificial intelligence knows before responding. An urban hotel with high weekday turnover does not need the same recommendation as a family resort in peak season. Nor should it interpret a cancellation, a rate drop, a late check-out request or a comment about breakfast in the same way. The question may be identical, but the hospitality meaning of the answer depends on context.

I have tested enough tools to distrust a misconception that is beginning to spread across our industry. We are treating the ability to converse with an artificial intelligence as though it were equivalent to having built intelligence into the hotel. It is not the same thing. ChatGPT can draft a report, summarise an incident or suggest a campaign. To do so well, it needs to know which hotel it is interpreting, what is happening today, which goals have been set, what promises have been made, what operational capacity exists and who can make each decision.

Nor is context a mountain of data supplied without judgement. A PMS may contain thousands of records and still fail to explain why a particular booking matters. A CRM may know a guest’s preferences without knowing which service is still available at ten o’clock at night. A Revenue report may show a drop in pickup without recording that nearby works have affected access to the hotel. Having information and having usable context are different capabilities, and confusing them leads to highly convincing responses that arrive dangerously late, recommend the unfeasible or overlook what truly matters.

That is where I see HotelGEX’s strategic value. Not merely in enabling a hotel to work through ChatGPT or Claude, but in helping those tools understand a specific hotel reality. The aim should not be for every hotel to use artificial intelligence, because that scenario will arrive sooner than we think. The aim is for every property to be able to bring to AI its own memory, rules, operational situation and business judgement. When the tool is shared, context is what begins to turn a common answer into a hotel’s own decision.

Profesional hotelero utilizando inteligencia artificial junto a información operativa y comercial del hotel

Artificial intelligence responds from the hotel it is able to know

A general-purpose artificial intelligence knows a great deal about Hospitality. It can explain what RevPAR is, propose a response to a review, design an email sequence or list measures to improve direct bookings. This knowledge is useful, but it is not enough to manage a real situation. Knowing how a hotel works does not mean knowing your hotel, just as reading hundreds of cookbooks does not tell you what is in a particular pantry today.

If you ask how to increase revenue next weekend, AI may recommend packages, upselling, segmented campaigns, length-of-stay restrictions or rate adjustments. These are reasonable possibilities. But does it know how many rooms are actually available? Does it know which room categories hold the inventory? Can it distinguish between low occupancy and limited operational capacity? Does it know whether the restaurant has space, the spa is fully booked, a group is under negotiation or a previous promotion has already damaged perceived rate value?

✦ HotelGEX
Guest Companion

Not another chatbot. An intelligence that stays with the guest.

From the first booking intent to the next stay, Guest Companion uses context to understand when to help, what may be relevant and when the best action is simply to stay silent.

✦ Guest Companion
Example journey
01
Discover Understand what kind of stay the guest is looking for.
02
Book Connect intent with the Booking Engine and real options.
✦
Prepare Arrival, preferences, relevant experiences and services.
04
Stay Useful help at the right moment — without unnecessary noise.
05
Return The next stay does not have to start from zero.
Understand · Anticipate · Recommend · Act · Stay silent — depending on context and timing.
Intelligence throughout the journey

Without those answers, AI works with industry knowledge, not business knowledge. The difference is decisive. The former produces plausible recommendations; the latter can produce applicable decisions. This distinction extends an idea I have already developed when stating that the smart hotel is the one that makes better decisions. To make better decisions, asking more questions is not enough. We need to build the context that makes it possible to interpret every answer.

In practice, I use a very simple test to assess the quality of an AI application in Hospitality. I call it the Interchangeable Hotel Test. It consists of removing the property’s name and replacing it with that of any other comparable hotel. If the response remains equally valid, we have probably obtained professional content, but we have not yet generated hotel-specific intelligence.

The test can be somewhat uncomfortable. Many AI-generated reports impress through their structure, tone and speed, even though they could be delivered unchanged to fifty properties. The same is true of commercial messages, procedures, complaint responses and hotel marketing plans. They are correct documents that contain too little local reality to alter a decision. Sometimes they even display that elegant solemnity which turns three ordinary ideas into seven pages and a meeting. In Hospitality, we had already accumulated enough experience of that before AI arrived.

To stop being interchangeable, an answer needs several layers of context that are often scattered across systems, documents, conversations and people:

  • Identity and positioning context. AI must understand what type of hotel it is interpreting, which guests it aims to serve, which commercial promise it upholds and which experiences are not part of its proposition. Without this layer, it will tend to recommend the same promotions, messages and services as it would to any other property, even where they contradict the positioning or attract poorly aligned demand.
  • Commercial and economic context. Occupancy, rate, inventory, pickup, channel, segment, acquisition cost, contribution and forecast demand alter the meaning of every recommendation. Selling more does not always mean accepting more bookings, and increasing revenue does not necessarily mean improving hotel profitability. AI needs to know which outcome is being protected and what cost may accompany the sale.
  • Operational context. A hotel cannot make promises solely according to what appears available in inventory. It must consider schedules, staffing, maintenance, food and beverage capacity, Housekeeping workload, incidents and the simultaneous delivery of services. A commercial recommendation disconnected from operations may increase revenue for a few minutes and generate compensation claims, strain and reputational damage for several days.
  • Guest and relationship context. Knowing the guest’s name, nationality or stay history is not enough. The reason for travel, the questions that preceded the purchase, the promises made and the needs affecting service all matter. As I explained when analysing why every booking can erase the conversation that made it possible, much of the commercial context disappears precisely when the guest becomes a confirmed booking.
  • Temporal context. Hotel information ages extremely quickly. Inventory changes, a fault is resolved, a group changes its dates, the weather alters an activity and a service ceases to be available. Data that was correct yesterday may lead to an incorrect response today. AI needs to know not only what it knows, but when it was true and how long it remains valid.
  • Authority and risk context. A recommendation should not automatically become an action. The hotel needs to define who can change a rate, grant compensation, contact a guest, share information or alter an operational routine. Artificial intelligence can propose; the organisation must decide what it can execute, what requires confirmation and what must remain expressly beyond its limits.

These layers explain why the same prompt can produce very different results. The phrase “prepare today’s operations report” appears sufficiently clear, but it actually contains many hidden decisions. What does the hotel consider a priority incident? Which deviation deserves attention? Which departments must act? What information is confidential? Which issue can wait and which one puts an imminent arrival at risk? Without these definitions, AI summarises data. With them, it can help manage the day.

Something similar happens with guest communications. Two hotels may ask ChatGPT to draft a pre-arrival message. The generic version will be friendly, well structured and probably too long. The contextual version will know what information that guest needs, what can be offered on that date, which services should be booked, which expectations need clarifying and which commercial opportunity makes sense. It will also respect the property’s tone without inventing attributes or amenities. That final risk is not insignificant, as I analysed when warning that AI can create a voice and promises that operations does not recognise.

Context also changes how reports are interpreted. A fall in ADR may appear negative until we discover that it stems from a deliberate strategy to increase average length of stay during low-demand dates. An increase in occupancy may be celebrated until it becomes clear that it has concentrated arrivals, breakfasts and departures into time slots the hotel can barely sustain. An improvement in reputation may conceal the fact that the highest-value guests are giving lower scores. Data describes what has happened; context helps explain what it means for that hotel.

This capability becomes even more important in chains, groups and hotel portfolios. Sharing a tool does not mean every property should receive the same recommendation. In fact, imposing a uniform artificial intelligence may digitally reproduce the same mistake that occurs when hotels in a chain try to improve by copying one another. Corporate criteria may be shared, but the response must recognise local demand, operations, market, team and constraints.

That is why I prefer to speak of hotel context capital. It is the organised body of knowledge that enables the property’s situation to be interpreted correctly and acted upon consistently. It includes data, but also rules, priorities, exceptions, relationships, memory and learning. It accumulates over time, can be lost when certain people leave and gains value when it becomes a shared capability.

Many hotels already possess this capital, even if they cannot use it systematically. It lives in the mind of the person who remembers why a decision was made, in the email that recorded a promise, in a spreadsheet understood by only one person, in the incident report, in booking notes and in conversations that do not reach the next shift. The problem is not merely a lack of information. It is the inability to bring it together around the decision that needs to be made.

Here I see an important contradiction. AI promises to democratise knowledge, but it can reinforce dependence on a few people if only they know what information must be provided to it. The hotel ends up using a new tool on an old organisational architecture. The experienced professional prepares the context, reviews the response and corrects errors; everyone else receives the result without understanding how it was built. We have gained speed, but not necessarily collective capability.

A sound hotel artificial intelligence strategy should reduce that dependence. Context must be accessible according to responsibilities, kept up to date and traceable. It must also allow the team to challenge the response. If a recommendation does not explain what information it used, what assumptions it made and what limits it observed, it is difficult to distinguish between a useful interpretation and a persuasive statement.

The quality of context can be assessed through six practical questions. They should be asked before trusting any AI-generated report, message or recommendation:

  • Is it relevant to the decision? Accumulating data does not improve the response if the important information is buried. Context must select what genuinely changes the judgement, rather than reproducing everything the system knows.
  • Is it up to date? Every element needs a recognisable period of validity. A schedule, rate, policy or availability should not continue to be used once it no longer reflects the operation.
  • Does it have an identifiable source? The team must be able to identify whether information comes from a booking, an approved policy, a conversation, an operating system or an inference. Its provenance determines the level of trust it deserves.
  • Does it distinguish facts from interpretations? The fact that a booking was cancelled is a fact; the assumption that it was cancelled because of price is a hypothesis. If AI mixes these two levels, it can turn an assumption into strategy.
  • Does it respect permissions and boundaries? Not every piece of available information should be included in every response. Context must protect privacy, confidentiality, security and decision-making rights.
  • Can it learn from the outcome? If the hotel accepts, corrects or rejects a recommendation, that experience should improve future context. Otherwise, AI will repeat the same errors with great efficiency.

When these questions are unresolved, the hotel risks confusing personalisation with assisted improvisation. The result may appear sophisticated, but it still depends on someone manually entering the right information in every conversation. This way of working provides individual usefulness, but it can hardly constitute a stable capability for hotel management.

HotelGEX turns context into a business capability

The contribution I find most compelling in HotelGEX begins with a simple idea. If the hotel wants to work through a conversational interface, the conversation must be able to draw on the reality of the property. The connection with ChatGPT or Claude opens an important door, as I explained when introducing the possibility for the hotel to work through conversational artificial intelligence environments. However, that connection gains value when there is sufficient hotel context behind it to respond accurately.

I do not believe HotelGEX should be understood as a collection of hotel prompts. A catalogue of instructions can save time and help people formulate better questions, but it still depends on each person adding the necessary information. Nor is it enough to connect a database and allow AI to query records. An isolated data point reports a figure; HotelGEX must help relate that figure to the hotel’s moment, objective, rules and capacity.

That nuance changes the nature of the product. Rather than using AI as a writer external to the business, the hotel begins to build a layer of understanding that supports reports, management tasks, communications and routines. Artificial intelligence no longer starts by asking who we are, what we want and what is happening. It can receive a coherent framework through which to interpret every task.

To explain this, I often use the concept of a Hotel Context Contract. It is not a legal contract, but an operational agreement on what a response must know before it can be considered valid. Each use case should define at least the following components:

  • The decision to be improved. “Analyse bookings” is too broad. It is better to specify whether the aim is to anticipate cancellations, identify upselling opportunities, review the profitability of a segment or prepare operations. The same information may require different interpretations depending on the purpose.
  • The authorised sources. The hotel must establish which systems, documents and records may be used, which prevails where discrepancies exist and who is responsible for their maintenance. Without a hierarchy of sources, AI may combine incompatible versions with admirable confidence.
  • The current operating status. The response must recognise inventory, occupancy, available services, incidents, expected pressure and temporary constraints. This layer prevents a good commercial idea from becoming a poor operational promise.
  • Business criteria. Context must include objectives, thresholds, priorities and exceptions. Not all hotels protect the same margin, pursue the same segments or accept the same level of risk.
  • Available authority. It is necessary to distinguish between informing, recommending, preparing an action and executing it. The greater the economic, reputational or personal impact, the more explicit human intervention must be.
  • The validity of the response. A recommendation may be useful for an hour, a day or a season. HotelGEX must help context expire once it is no longer reliable, rather than allowing an old response to continue circulating as though it were current policy.
  • Subsequent learning. The system needs to capture what happened afterwards. If an action increased conversion, created operational workload, generated a complaint or was rejected by the team, that outcome should inform future decisions.

With this contract, AI use is no longer measured by the number of conversations opened and begins to be assessed by the quality of the decisions changed. This transition is fundamental. A hotel may generate hundreds of texts every month and fail to improve a single experience. It may also use HotelGEX for a few critical decisions and achieve a noticeable impact on sales, coordination and satisfaction.

In reporting, for example, HotelGEX can help move from summary to interpretation. It would not simply indicate occupancy, arrivals, departures and incidents, but could relate these variables to the day’s operational pressure, commitments made and pending decisions. The report would cease to be an administrative snapshot and become a prioritisation tool.

In hotel Revenue Management, context helps avoid recommendations that observe demand and rate without considering the cost of servicing it. A revenue opportunity may require a particular room category, concentrate included services, extend opening hours or displace more profitable demand. AI should be able to show these tensions, rather than merely suggesting higher or lower prices.

In hotel marketing, HotelGEX can help ensure communication is born from what the property can genuinely deliver. The campaign should recognise availability, compatible segments, active services, seasons, restrictions and the brand promise. In this way, AI not only writes more accurately; it reduces the gap between what Marketing promotes and what the team will have to deliver.

In the hotel guest experience, context turns a standard response into continuity in the relationship. If the guest has already explained the reason for their trip, requested a particular condition or received a promise, they should not have to start again each time they change channel or department. Technology adds value when it avoids repetition and enables the hotel to act as one organisation.

In daily operations, the difference can be seen in seemingly small matters. A late check-out request should not be answered merely by checking whether the room is vacant. The next arrival, Housekeeping workload, cleaning time, guest value, contracted conditions and the precedent that an exception may create also matter. Context helps resolve with judgement what would otherwise end up depending on mood, memory or whoever happens to be at Reception.

In leadership and coordination, HotelGEX can make it easier for teams to work from a shared interpretation. The same data should not become five departmental versions. Revenue may see an opportunity, Operations an overload, Finance an insufficient margin and Marketing a campaign. Intelligence emerges when these perspectives can come together around the same decision without erasing their tensions.

For this capability to work, the hotel will need to resist the temptation to contextualise everything from day one. I have seen too many projects attempt to reproduce the entire organisation before improving a single decision. The result is usually an impressive architecture, months of configuration and some difficulty explaining which problem has disappeared. I recommend progressing through a far more hotel-oriented and less ceremonial sequence:

  1. Choose a frequent and economically relevant decision. It may be preparing the daily report, managing arrivals, responding to requests, identifying commercial opportunities or reviewing bookings. It must be specific enough for improvement to be observed.
  2. Identify the minimum necessary context. It is worth asking what information an experienced person uses today to make that decision. This observation often reveals rules and exceptions that never appeared in the manuals.
  3. Separate stable information from changing information. The hotel’s identity, positioning and certain policies may change slowly. Inventory, schedules, incidents and availability require much more frequent updates.
  4. Define sources, owners and expiry. Every contextual element must have a source, a responsible person or role, and a validity criterion. If no one can confirm a piece of information, AI should show uncertainty rather than fill the gap.
  5. Set AI authority. The hotel must decide whether HotelGEX may consult, summarise, recommend, prepare messages or trigger actions. Autonomy should increase only when the quality of context and the risk of the decision allow it.
  6. Compare the response with human judgement. During implementation, it is useful to ask experienced professionals to explain what is missing, what is unnecessary and which interpretation they consider incorrect. This comparison turns tacit knowledge into shared context.
  7. Measure the outcome, not just time savings. A fast response may be irrelevant or wrong. It is worth observing whether errors, rework, broken promises and repetition decline, and whether conversion, margin, coordination and satisfaction improve.

I also propose measuring an indicator I call the Replaceable Response Rate. It consists of reviewing a sample of generated responses and estimating how many could be used unchanged at another hotel. The higher that proportion, the less differentiated context has been incorporated. Not every response needs to be unique, of course. A translation or an administrative summary may be perfectly generic. However, commercial, operational and relationship recommendations should clearly reflect the property receiving them.

To this metric, I would add the Time to Verified Response. What matters is not merely how long AI takes to respond, but how much time a person needs to check, correct and complete the result before using it. A report generated in twenty seconds that requires half an hour of review has not truly reduced the work; it has merely moved it to a less visible stage.

Another useful measure is the Contextual Correction Rate, which records how many responses must be amended because information was missing, outdated data was used or an incorrect rule was applied. This metric helps identify where the system is weakening. Sometimes the problem lies in AI; quite often it lies in the fact that the hotel had never clearly defined its own judgement.

That discovery may be one of HotelGEX’s most valuable contributions. In attempting to teach the hotel to an artificial intelligence, the organisation discovers everything that existed only in the minds of certain people, everything different departments interpret differently and every policy that functions through undocumented exceptions. AI then acts as a mirror of operational maturity.

It is wise to retain a degree of modesty about this possibility. HotelGEX cannot create on its own a context the hotel does not possess, nor resolve contradictions the organisation refuses to confront. If Sales promises something that Operations rejects, if Revenue and Finance calculate profitability differently, or if every shift applies its own policy, connecting those data points will not automatically produce a shared truth. It can, however, make the discrepancy visible and compel us to decide.

This is an important difference from conventional automation. An automated process repeats a sequence. A contextual capability must interpret change, recognise exceptions and show uncertainty. In a hotel, where two days with the same occupancy can generate entirely different pressures, this sensitivity matters more than raw speed.

It also requires caution with personalisation. The fact that HotelGEX can use more information does not mean it should display all of it or turn every known detail into a visible intervention. Guests appreciate the hotel remembering what is useful, but they may feel watched when they discover that we retain information they did not expect. Context must improve service without turning hospitality into surveillance.

The same caution applies to teams. A contextual tool can facilitate productivity reports, incident tracking and task allocation, but it should not be used to create opaque assessments of people. Leadership in the hotel sector needs to explain what information is used, for what purpose and which decisions should never be delegated to an automated interpretation.

When well governed, context delivers advantages across several dimensions at once. It improves the quality of responses, reduces the work of explaining the hotel’s situation repeatedly, strengthens consistency between departments, protects the commercial promise and allows learning to survive changes of shift or personnel. It also turns operational knowledge into a less fragile asset.

From a business perspective, this capability can help sell more by identifying opportunities compatible with inventory, guest and timing. It can help sell better because it considers contribution, costs and capacity before recommending an action. And it can improve satisfaction because it enables continuity, precision and less effort for the guest. More sales, better margin and happier guests are not three independent goals when context connects the complete decision.

If you are considering how to introduce HotelGEX into your property, my advice is not to begin by asking how many tasks artificial intelligence can perform. Choose a decision that currently depends too heavily on searches, memory, interruptions or specific people. Describe what a competent professional needs to know to resolve it, and check whether that knowledge is available, up to date and authorised. This exercise provides a far more useful starting point than any feature catalogue.

HotelGEX can become a competitive differentiator precisely because it does not seek to make every hotel think alike. Its value emerges when it helps each property bring its own reality to ChatGPT or Claude and receive a response that recognises its guests, operations, limits and objectives. The tool may be shared; the understanding of the hotel should not be.

Before long, saying that a hotel uses artificial intelligence will have much the same value as saying that it uses email. The professional question will be different: what does that AI know about your property that enables it to help you make better decisions? Start with one specific routine, protect the quality of the context and measure what changes afterwards. The distance between an elegant response and genuine improvement is usually found precisely there.

KEEP EXPLORING

This article ends here. The archive does not.

Lead Hospitality brings together 1200 English articles published since 2018: years of experiences, decisions and lessons you can keep exploring.

CONTINUE FROM HERE

What would you like to explore next?

Choose a direction and keep reading around what you want to solve, learn or challenge.

Discover something interesting ↓Surprise me ↗
01MAKE MOREProfitability, revenue and decisions that reach the bottom line. 02LEAD BETTERTeams, culture, talent and the conversations that matter. 03SELL BETTERPositioning, marketing, distribution and customers. 04CREATE EXPERIENCESService, loyalty and details guests remember. 05UNDERSTAND WHAT'S NEXTInnovation, AI and new ways of thinking about Hospitality. ✦SURPRISE MEShow me something worth five minutes of my time.
✦ LEAD AI · MEMBERSHIP

Turn reading into actionable knowledge

Basic includes article recommendations. Premium unlocks questions, analysis, comparisons and applying the knowledge to your hotel.

See subscriptions →