The arrival of new AI models has accelerated the automation and optimisation of processes, calculations, rates and big data analysis in hotel businesses, and those who have managed to incorporate them naturally into their daily workflow share an important conclusion: human oversight guided by professional judgement is key. No one with genuine responsibility over a P&L — least of all someone who works with these tools every day and knows first-hand where they fall short — hands over areas as sensitive as Revenue Management or SEM marketing campaigns to an AI without keeping their own judgement inside the decision loop.
Confusing the computational power AI offers with the judgement and leadership that day-to-day management demands is a categorical error, not a matter of nuance, and it costs you in budget and in ADR. A machine’s processing capacity cannot manage the commercial and operational complexity — online and offline — of a hotel business. The more you work with these tools, the less you see them as replacements and the more you value them as amplifiers of human capability. The right stance is the one held by the hybrid teams that work with it best: AI is an assistant that expands analytical capacity and frees up hours of mechanical work, while professional judgement remains the only legitimate decision layer — giving managers in each area greater control, faster analysis and sharper execution. Everything else is sales talk that sooner or later runs into the reality of daily operations.
Executing is not deciding
We can all agree that no person can evaluate thousands of signals in milliseconds, or recalculate each night the price curve for three hundred and sixty-five dates cross-referencing occupancy, pickup, compset and seasonality. On that ground there is no logical argument to be made, and claiming otherwise would be absurd. The problem arises when the commercial narrative extends that technical competence into the decision layer — where you define what constitutes a value conversion, which segment you want to grow and which to reduce, which source markets justify which acquisition cost, how much cannibalisation you accept between direct channel and intermediated distribution, and which price protects positioning even if it penalises this week’s occupancy.
None of those questions can be answered inside Google Ads or inside an RMS, to name two examples. The system does not know that your target ADR in low season responds to a repositioning strategy and not to a need to fill rooms, it does not know that the corporate segment with the worst conversion rate is the one that sustains your February Mondays, it does not know that a seven-night booking with breakfast is worth more than two three-night stays with the same aggregate amount because the operational cost per arrival is not in its model. When someone claims that AI manages online marketing campaigns or Revenue Management on its own, they are making a mistake born of naivety and ignorance of the industry — one that ends up being paid for in operations, and also in the bottom line…
What the pricing engine cannot know
The forecast of any RMS is built on its history. For example, a property with two years of data distorted by a renovation, a management change, a product repositioning or a structural shift in its source market produces forecasts that carry that distortion forward with impeccable mathematical conviction. The engine does not know that the history it is reading does not describe the hotel it manages today. Only the Revenue Manager who was there knows that, and that asymmetry of context is precisely the value a professional brings compared to a software licence.
The problem repeats itself with uncaptured demand: the system observes materialised demand, not the demand lost to price, restriction or availability closure, and therefore tends to reinforce its own previous decisions. If you closed a date too early last year, the historical data records full occupancy at a given price and the algorithm concludes that price was the right one. No one inside the model can know how much you left on the table. That analysis requires reading denied demand, regret data and the pattern of unconfirmed requests with interpretive judgement, and it requires someone who remembers what actually happened that week. The same applies to the compset: a poorly defined rate shopping setup optimises against the wrong benchmark with absolute efficiency, and the engine cannot detect that the hotel across the street has shifted segment after a renovation, joined a chain or is selling below cost to build its reputation at opening.
The black box: the less you see, the more judgement matters
The case for human oversight of AI has a prerequisite that many platforms are quietly eroding: applying judgement requires information. A Revenue Manager can challenge a pricing recommendation because they understand what data underpins it, what assumptions lie behind it and what its potential limitations are. Without visibility, applying judgement does not become harder; it simply becomes impossible. Something similar happens in digital advertising management. Systems like Performance Max, AI Max or automated bidding concentrate ever-greater decision-making capacity inside algorithms whose actual workings are opaque to the user. The problem is not merely one of transparency or reporting. The problem is that when the capacity to observe, interpret and challenge disappears, so does much of the capacity to lead.
It would, however, be a mistake to conclude that this opacity makes artificial intelligence an unquestionable authority. In fact, exactly the opposite is true. The more complex and less visible the system, the greater the need for professionals capable of interpreting results, detecting anomalies, understanding the business context and deciding when to follow a recommendation and when to ignore it. AI can optimise thousands of variables simultaneously, but it does not understand the organisation’s strategic objectives, the hotel’s operational constraints, shifts in the local market or the commercial decisions that have yet to appear in any data. This is where one of the most common mistakes in AI adoption appears: assuming that greater computing power implies greater decision-making capacity. They are different things. The first belongs to the machine; the second still belongs to people. Believing that an algorithm can manage SEM on its own reflects a naïve understanding of how this works.
The value of alignment between marketing and Revenue Management teams, beyond AI
Within a single hotel operation, two systems coexist with different objective functions. The marketing one optimises conversion volume and gross booking value. The Revenue Management one optimises price by date, seeking to maximise revenue per available room. Nobody has aligned them because no technical layer does so, and the result is a hotel spending money to acquire demand on dates that were already in compression — displacing higher-value bookings that would have come in without advertising spend — while leaving the low-demand dates without support, where every euro invested in acquisition would have generated real incremental return.
The ROAS of that campaign can be spectacular and be precisely the problem at the same time. The platform reports a high return because it attributes bookings that would have happened anyway, and the hotel celebrates an efficiency that, in terms of net contribution, does not exist. Detecting this requires cross-referencing the campaign report with pickup by stay date and with the forecast — not the click date — and it requires someone with a revenue mindset to read the marketing data and someone with a marketing mindset to read the revenue data. That conversation is 100% human, and no integration replaces it, because the problem is not one of data connectivity but of alignment between your marketing and revenue teams.
The model that actually works: the machine proposes, judgement decides
None of the above leads to rejecting technology, which would be a position as indefensible as the opposite one. It leads to placing it where it performs. AI is extraordinary at calculating scenarios, detecting demand anomalies before the human eye catches them, cleaning and cross-referencing volumes of data no analyst could process, generating creative variants to test and freeing the team from the mechanical part of the job. That is its role: a high-performance assistant that expands the capacity of whoever decides. The professional contributes what no model has: context, business memory, market reading, commercial judgement and accountability for the outcome.
In practice, this translates into a concrete method. The engine proposes a rate and the Revenue Manager validates it against what they know about the real compset, the events calendar that has not yet shown up in observed demand, the positioning they want to protect and the mix they want to build. The platform proposes a bid and the marketing team validates it against the forecast, protects brand traffic, decides which dates merit incremental investment and which do not need it. Automation executes thousands of operations per second within limits a human has defined, with alert thresholds a human has calibrated, and with periodic review to ensure the system is still optimising what it was asked to optimise. None of that is technological distrust; it is data governance.
The tasks that cannot be delegated are precisely those that determine profitability: defining what a value conversion is, auditing the integrity of the circuit between PMS, booking engine and platforms, importing net values, setting thresholds before activating any automation, monitoring the real composition of traffic and mix, holding the price when the algorithm pushes to lower it and always cross-checking reported return against realised revenue. These are not execution tasks; they are decisions, and decisions require context, expertise and someone who is accountable for them. No one who has managed a budget with an obligation to justify its profitability would argue otherwise. Without that judgement, AI is not working for your hotel: it is working for the platform that sells it to you, and it does so with remarkable efficiency.
