By Michael McCartan, Area Vice President EMEA at IDeaS
Hotels are investing significant time and resources in artificial intelligence, yet many are still struggling to move from experimentation to widespread operational use.
The challenge facing many hotel groups today is no longer access to AI. It is developing enough confidence in the technology, data and governance behind it to allow AI to influence real commercial decisions at scale.
As a result, many organisations find themselves trapped in a form of “AI pilot purgatory”, where teams are willing to test the technology and review its recommendations but remain hesitant to let it influence live commercial decisions.
The trust gap becomes particularly visible in revenue management, where changes to pricing, inventory or forecasting can have an immediate financial impact.
McKinsey’s State of AI 2025 found that fewer than one-third of organisations have successfully scaled AI beyond the pilot stage. For hotels, the challenge is often not whether AI works. It is whether organisations trust it enough to become part of everyday decision-making.
COMMERCIAL DECISIONS RAISE THE STAKES FOR TRUST
Trust becomes more important as technology moves from providing insight to influencing decisions.
A revenue manager may be comfortable reviewing a recommendation generated by a system. Allowing that same system to automatically increase a rate, restrict lower-rated inventory or decline a piece of business requires a higher level of confidence in both the data and the decision-making process behind the recommendation.
Revenue management systems often reach conclusions that appear counterintuitive when viewed through a single data point such as current occupancy. A hotel may still have rooms to sell while forward demand, booking pace, length of stay and displacement indicators suggest that restricting lower-value business will produce a stronger commercial outcome.
Teams therefore need sufficient visibility into how a recommendation was reached, what factors influenced it and where human intervention remains appropriate. Trust develops through consistent performance, but it also depends on understanding.
Accuracy alone is not enough. Commercial teams are more willing to adopt automation when they can understand the factors that influenced a recommendation and communicate that reasoning across the business. Trust is strengthened when AI can be governed as well as measured.
DISCONNECTED SYSTEMS LIMIT WHAT AI CAN ACHIEVE
Technology is only as effective as the information available to it.
Hotels typically manage commercial data across a range of systems, including the PMS, CRS, RMS, CRM and distribution platforms. The quality of any automated recommendation depends heavily on how consistently those systems exchange information.
Disconnected data can limit what AI is able to achieve, while inconsistent reporting and definitions can cause different departments to make decisions using different versions of demand.
Siloed ways of working create additional challenges. Revenue teams may be reacting to booking pace, marketing teams may be focused on generating demand and sales teams may be pursuing group business, all while operating from different assumptions about future performance.
Introducing AI into that environment can increase the speed of activity without necessarily improving the quality of decisions if the information feeding those decisions remains incomplete or fragmented.
Applying AI to fragmented data does not simply risk producing a poor recommendation. When the wrong type of AI is paired with incomplete or disconnected information, organisations risk creating bad recommendations at scale.
Instead of eliminating manual inefficiency, they may simply replace it with automated error-making. Recommendations can be repeated confidently across room types, market segments and future arrival dates, multiplying the impact of flawed assumptions or incomplete inputs.
The danger is that the system may still appear to be working. Without the right architecture, monitoring and explainability, errors can remain hidden until they have already influenced a meaningful volume of commercial activity.
That is why hotels increasingly need to understand not only what data feeds an AI system, but also what form of AI is being used, where it sits within the decision process and how its recommendations can be evaluated and challenged.
HOTELS ARE STARTING TO ASK A DIFFERENT QUESTION
For several years, much of the hospitality industry’s conversation around AI focused on whether technology providers offered AI capabilities.
Increasingly, however, hotel groups are asking a different question: how deeply does AI actually sit within the decision-making process?
The question matters because not all AI-enabled systems are designed in the same way. Some technologies use AI to summarise information, generate content or surface insights while core commercial recommendations continue to rely on predefined rules or manually maintained logic.
Others place machine learning at the centre of forecasting and optimisation, allowing models to continuously adapt as demand conditions change.
As hotels pursue greater automation and scalability, the underlying architecture becomes increasingly important. Trust is influenced not only by the quality of a recommendation but also by confidence that the decision process is consistent, explainable and embedded within the product itself.
This reflects a broader shift in how hospitality organisations evaluate AI. The question is no longer simply whether AI exists within a platform. Increasingly, buyers want to understand where AI sits within the decision process and whether it is capable of acting as part of the commercial decision engine itself.
HOTELS NEED TO MATCH THE TECHNOLOGY TO THE DECISION
The rapid growth of generative AI has created a tendency to discuss AI as a single category, even though different commercial challenges require different forms of intelligence.
Generative AI can support communication, interpretation and user interaction. Revenue management relies on mathematical forecasting and optimisation to estimate future demand and determine how pricing and inventory should respond.
A generative model predicts language. A revenue management system must forecast demand across future arrival dates and optimise commercial decisions based on those forecasts.
Mathematical AI can support the decision engine by collecting demand signals, forecasting future behaviour and optimising pricing and inventory. Generative AI can help users interpret and communicate recommendations, while agentic capabilities can support execution by carrying out approved actions within defined parameters.
The most effective commercial environments combine these capabilities within a single decision framework. Forecasts, recommendations, explanations and actions become more valuable when they work together as part of a connected system rather than as separate tools.
When the wrong type of AI is asked to solve the wrong problem, organisations risk creating confidence in outputs without improving the quality of the underlying decision.
The commercial value comes from creating a continuous process in which signals are collected, demand is forecast, controls are optimised, decisions are applied and future recommendations improve through ongoing learning.
HOTELS NEED TO DEFINE WHERE AUTOMATION STOPS
Trust also requires clear accountability.
Hotels need to establish which decisions can be automated, which circumstances require review and who remains responsible for outcomes.
Routine pricing adjustments may operate automatically within agreed parameters, while unusual demand shifts, major deviations from forecast or significant inventory decisions may warrant additional oversight.
Clear intervention points help teams understand where human judgement remains necessary and make it easier for automation to become part of everyday operations rather than remain confined to controlled pilot programmes.
Governance must also be supported by skills, training and adoption. Technology alone does not create confidence. Employees need to understand why systems are being introduced, what role automation will play and how decision-making responsibilities evolve as AI becomes more deeply embedded within commercial processes.
SUCCESSFUL ADOPTION STARTS WITH A SPECIFIC PROBLEM
Hotels can prolong pilot purgatory when AI initiatives begin with a broad ambition to adopt AI rather than a clearly defined business challenge.
A more effective starting point is usually a specific outcome, such as improving forecast accuracy, responding more quickly to changes in booking pace, reducing manual pricing activity or evaluating group opportunities more consistently.
Once the objective is clear, organisations can determine what data is required, which technologies are appropriate and how success will be measured. They can also establish the evidence needed to justify moving from experimentation into operational use.
Moving beyond pilot purgatory ultimately requires organisations to think about AI less as a technology project and more as an operating model.
The discussion is no longer simply whether AI can generate recommendations. The real question is whether organisations trust the underlying decision engine enough to embed those recommendations into everyday commercial processes.
When that happens, AI stops being an experiment and becomes part of how the business operates.