Over the following days, I kept thinking about a distinction that had appeared almost at the end of the previous conversation. We had established that it was possible to reconstruct part of the hotel’s identity using only signals available from the outside, and that some of those signals acquired a different value when we stopped asking what made the property special and began asking instead for whom it might be particularly well suited.
But that conclusion had left open a question we had not yet tried to answer:
How much could I trust what I had just understood?
I returned to the same sources. The website still said that the hotel offered personalised attention. Similar expressions appeared in different places, and some booking platforms reproduced descriptions that seemed to confirm it. There were photographs capable of suggesting care, policies that contemplated certain particular situations, and several reviews in which guests remembered adaptations, names or decisions that did not seem to reflect completely standardised attention. If my only objective had been to gather signals favourable to that characteristic, I could have found enough to fill several pages.
Yet something was beginning to trouble me. Many of those signals were not as independent as they appeared.
A description published on a booking platform might come from the same text the hotel had written for its own website. Another platform might have taken information from an earlier source. A press article might reproduce almost word for word a presentation supplied by the property itself, while several social media posts might simply be different pieces from the same campaign. Counted as appearances, they seemed numerous. Traced back to their origin, some were nothing more than the same claim repeating itself in different places.
Until then, we had been looking at the presence of signals and the compatibility between them. I was beginning to understand that two pieces could agree perfectly without providing two separate reasons to believe what they were saying.
I thought of a fairly common situation. A hotel describes itself as particularly quiet. Its website uses that word, a booking platform incorporates it into the description, a tourism publication repeats it, and a social media account mentions it again. Four sources appear to lead towards the same conclusion. But if all four originated in a single description written by the hotel itself, perhaps we do not have four confirmations.
Perhaps we have a single claim that has learned to travel.
The idea became even clearer when I returned to the reviews. Repetition appeared there too, but of a different kind. People who seemed to have no connection with one another, who had travelled at different times and described different situations, ended up remembering something similar. One spoke of tranquillity at night; another mentioned barely hearing any noise from the room; a third explained that they had been able to rest after a long journey. None used exactly the same wording and perhaps that was precisely what made the agreement more interesting. There seemed to be no original phrase that the others were reproducing. Different experiences were leading, each by its own path, towards a similar conclusion.
I began to look at signals differently. I was no longer interested only in how many I could find, but in how much new knowledge each one contributed. A characteristic mentioned twenty times might be supported by a single original source, while another appearing only four times might come from four independent experiences. Counting them produced one result; asking where they came from produced quite another.
I continued through the reviews and encountered a different difficulty. I found one particularly enthusiastic review in which a guest described in considerable detail how the team had resolved an exceptional situation. They had arrived under unforeseen circumstances, needed an adaptation that was not obviously provided for, and someone had found a solution. The scene fitted almost perfectly with the logic we had been observing for several conversations: first understand the person; then decide how to look after them.
For a few seconds, I was tempted to treat it as particularly strong confirmation. It was specific, described in detail and came from someone who had lived through the experience. But a single question was enough to temper some of my enthusiasm:
Did this describe the hotel, or did it describe what happened once?
There was no reason to doubt the guest. The episode could have happened exactly as described and still not demonstrate that anyone in a similar situation would receive a comparable response. Perhaps someone particularly resourceful happened to be working that day. Perhaps exceptional circumstances had made the adaptation possible. It might even have been standard practice, but a single experience was not yet enough to know.
The specificity of a signal helped us understand more clearly what had happened without necessarily turning that episode into a rule.
That distinction stayed with me throughout the rest of the review. General claims made by the hotel could express a genuine intention without demonstrating its execution. Policies showed what possibilities had been anticipated, though not necessarily how they worked when someone tried to use them. Photographs were extraordinarily convincing for certain visible characteristics and far less useful for others. Reviews brought us closer to real experiences, but each still described a particular stay shaped by specific expectations, circumstances and people.
A source could be extraordinarily reliable for one question and almost useless for another.
That also dismantled the temptation to arrange sources into a fixed hierarchy, trusting everything the hotel said less and everything its guests reported more. If I needed to know whether a room measured thirty-two square metres, I would probably trust an up-to-date technical specification from the property more than a guest’s impression that the room was enormous. If I wanted to know the current breakfast hours, an official policy might be more useful than a review written two years earlier. If, on the other hand, I was trying to understand how the team responded to an unexpected arrival, the institutional description might tell me less than several recent experiences from people who had faced precisely that situation.
The question was no longer which source was more reliable, but something much more precise:
Reliable for what?
That small change made the analysis considerably more complicated, but at the same time much closer to the way we make decisions outside hotels. We do not use the same evidence to check the temperature of a room as we do to judge someone’s friendliness. Nor do we expect to establish a cancellation policy in the same way that we try to understand the atmosphere of a place. Each claim seemed to require signals capable of getting reasonably close to what we were trying to know.
I returned to the photographs with that idea in mind. An image could show with great precision that a room had a desk, a bathtub or a particular arrangement of beds. It could also help us understand its brightness or its relationship with the landscape, although framing, time of day and editing were already beginning to play a role there. It was far more difficult to use that same photograph to support the claim that the hotel offered personalised attention. It might suggest care, perhaps even warmth, but the distance between what was visible in the image and the characteristic we wanted to establish was too great.
I began to recognise that distance almost everywhere. Some signals were very close to what we were trying to understand; others required several layers of interpretation before we could get there. A policy explicitly explaining how to coordinate a late arrival came fairly close to establishing that the possibility existed. A photograph of a welcoming reception area could be much further from establishing flexibility in response to an unexpected arrival, however much both might contribute to a general impression of hospitality.
The more interpretation we needed to add ourselves, the more cautious our conclusion needed to be.
One of the reviews we had used several times showed exactly that difference. The guest did not simply say that the service had been excellent. They described what they had needed, what had happened and how the team had responded. Excellent service might be a completely sincere opinion, but it required us to imagine what excellent meant to that person. A concrete scene reduced some of that distance because it allowed us to observe the event on which the guest had based their assessment.
Something similar happened with the words used by the hotel itself. We are flexible required us to believe in a characteristic. If your flight arrives after the usual time, let us know and we will coordinate your arrival described a verifiable possibility. Both statements could come from exactly the same property and be equally true, but they did not provide the same kind of information.
I was beginning to understand why, during the previous conversation, certain small operational details had proved more revealing than much more ambitious claims. It was not necessarily because they said something more important, but because what they said could be checked more easily.
I continued and encountered another problem. Several of the signals that seemed strong had a date attached to them. A detailed review about a late arrival had been written four years earlier. A photograph showed a room before a renovation. A post mentioned a service that was no longer listed as available, while a response from the hotel described a policy that appeared to have changed.
None of those pieces became false simply because they had aged. They still said something true about the property that had existed at that time. The difficulty was different: we were trying to decide whether that information allowed us to understand the hotel someone could choose today.
For the first time, recency stopped seeming like a matter of digital maintenance and became part of trust.
A signal could be independent, specific and extraordinarily concrete, but if it described a reality that no longer existed, it could lead to the wrong recommendation. At the same time, it did not seem reasonable to automatically discard everything old. A characteristic remembered consistently over many years might reveal something much deeper than a novelty mentioned for only a few weeks. Age could weaken operational information and, in another context, strengthen the idea that a particular characteristic had endured over time.
I wanted to see what happened when those differences entered the same interpretive process we had used in the previous conversation. I returned to the artificial intelligence system and presented it with some of the signals we ourselves had found, modifying them little by little to observe what happened. If several descriptions repeated the same claim, the conclusion appeared to strengthen. If I added specific, recent reviews from different guests, it gained even greater consistency. If I introduced a policy that contradicted that characteristic or recent comments describing opposite experiences, the response began to become more cautious.
I was not interested in determining whether that particular system assigned the correct weight to each source. That would have shifted the conversation towards the workings of a particular tool. What mattered was something else: to recommend with a certain degree of precision, any process of interpretation needed to do more than find information. It had to decide how much to trust it.
Until then, we had been thinking mainly about helping the hotel to be understood. We had discovered a possible identity, looked for its traces, tried to reconstruct it from the outside and observed how certain characteristics could become important in relation to specific needs. But if we now wanted those characteristics to play a part in a recommendation, a new responsibility appeared.
Making a claim visible was not enough. Getting many pages to repeat it was not enough either. Even accumulating favourable reviews could prove insufficient if we did not know exactly what they were confirming.
Trust seemed to be built in a less spectacular way. A claim found a decision that made it possible. A policy showed that the decision did not depend entirely on improvisation. Different guests described having experienced something compatible. The signals were recent enough to continue describing the property as it existed today and, when they came from different places, did not appear merely to repeat the same original source.
No single piece demonstrated the complete identity, but together they progressively reduced the number of things we needed to assume.
Perhaps that was where a more useful definition of trust began to emerge:
reaching a conclusion while needing to add less and less of our own.
I returned then to the formulation that had organised the previous conversations: first understand the person; then decide how to look after them. If we found it only written on the homepage, we knew that the hotel wanted to present itself that way. If we also discovered policies that allowed for adaptations, operational situations that showed responsiveness and different guests who spontaneously remembered being looked after according to their circumstances, we began to have something different.
Not because we had accumulated a particular number of proofs, but because independent paths were beginning to lead towards the same place.
Yet as I tried to organise those signals, something appeared that we had not considered sufficiently. Among the recent reviews, I found one describing a different experience. The guest had arrived outside the usual hours and reported difficulties coordinating entry. It did not negate everything that had come before. Nor did it demonstrate that the hotel was rigid or make the positive experiences we had read false. But it was there, it came from a situation directly related to what we were trying to understand, and it could not simply disappear because it was inconvenient for our conclusion.
Until then, we had learned to look for signals that agreed. Now one had appeared that did not, and my first reaction was to try to explain it. Perhaps there had been a communication problem, an exceptional circumstance, a change of shift or an expectation the property had never promised to meet. All were reasonable possibilities, but none was contained in the information available.
To preserve our interpretation intact, we would have to begin inventing what we did not know—exactly the opposite of what we had been trying to do.
I closed the page without resolving the contradiction. The signal remained there alongside all the others: a policy suggesting flexibility, several experiences that appeared to confirm it and another, sufficiently specific and recent, showing something different.
Learning to trust signals, I realised, would not consist solely of discovering which were stronger. We would also have to decide what to do when equally credible signals stopped leading towards the same conclusion.
Until then, coherence had helped us understand the hotel.
Contradiction was about to show us how much we could trust that understanding.
Management Perspective
Not all signals allow us to understand a characteristic with the same degree of confidence. Some are closer to what we are trying to know; others require additional interpretation. Some provide an observation of their own, while others reproduce information originating elsewhere. A claim may express an intention, a policy show that a particular possibility exists, and a concrete experience reveal how it was lived. When sufficiently different, current signals that are close to what we want to understand lead towards a compatible conclusion, we need to add fewer assumptions in order to sustain it. That is precisely where trust begins to be built.
The Case Study
In the hotel we have been observing, we stopped counting signals as though they all contributed the same knowledge. Some appearances came from a single source; certain concrete experiences showed what had happened without yet demonstrating that it was standard practice; a policy proved highly reliable for some questions and of little use for others; and some signals had aged enough that they could no longer describe the current property with confidence.
When different sources began to lead, along their own paths, towards a similar conclusion, our confidence increased.
Until a sufficiently specific and recent experience told a different story.
A Question Before Moving On
If you found ten signals confirming what your hotel says it is and just one—recent, specific and difficult to ignore—showing something different, what would you need to know before deciding which of those stories better describes the experience you are offering?