A Hotel CIO on Building an AI Strategy That Actually Works

When Marcus Devereaux became Chief Information Officer of a 120-property European hotel group three years ago, his brief was deceptively simple: modernise the technology estate. What he discovered was that the technology was rarely the problem. The problem was the organisation's relationship with data, and until that changed, no amount of new software would move the needle.

We spoke with Marcus about what it really takes to build an AI strategy in a mid-size hotel group, the politics, the infrastructure requirements, the vendor landscape, and the moments where the whole thing nearly fell apart.

Where Most Hotel AI Strategies Go Wrong

**Where do most hotel AI strategies go wrong?**

They start with the use case instead of the data. Hotels will say, "We want to use AI to personalise the guest experience," and immediately start evaluating vendors. But personalisation at scale requires clean, unified guest data across every touchpoint, check-in, F&B, spa, loyalty, OTA bookings, direct bookings. If that data lives in six different systems that don't talk to each other, you don't have a personalisation problem. You have a data infrastructure problem. AI cannot fix that.

Building the Data Foundation First

**How did you approach the data infrastructure challenge?**

We spent eighteen months on what we called the foundation layer before we touched a single AI use case. That meant building a customer data platform that ingested from our PMS, CRS, POS, and loyalty systems. It meant establishing data governance, who owns which data, what the quality standards are, how we handle consent and privacy. It was unglamorous work. Nobody was presenting it to the board with beautiful slides. But it was the work that made everything else possible.

The First AI Deployment: Revenue Management

**What was the first AI application you deployed?**

Revenue management. It was the easiest case to make internally because the ROI is direct and measurable. We replaced a traditional rules-based pricing system with a machine learning model that incorporates over 200 signals, competitor rates, local events, weather, demand patterns, booking pace. Within six months, RevPAR was up 4.2% against the competitive set. That result gave us the credibility to pursue more ambitious applications.

Winning Organisational Trust for AI

**What did the organisation struggle with most?**

Trust. Revenue managers who had spent twenty years building their instincts didn't want to hand pricing decisions to an algorithm. And honestly, they shouldn't, not entirely. The model we deployed requires human oversight. It recommends; humans decide. What changed was that the recommendations were so consistently good that the revenue managers started spending their time on exceptions and strategy rather than rate management. That's the right dynamic.

Building that trust took time and deliberate communication. We were transparent with the teams about how the models worked, what data they were trained on, and where they were likely to make mistakes. When the AI got something wrong, and it did, we used those moments as learning opportunities rather than crises. Transparency about limitations turned out to be as important as demonstrating performance.

The Next Frontier: Operational AI

**Where is the technology going over the next three years?**

The frontier right now is operational AI, using computer vision and sensor data to optimise housekeeping, maintenance, and energy management. We have pilots running in four properties that are reducing energy consumption by 12-15% with no guest experience impact. The technology is proven. The challenge is deploying it at scale without disrupting operations. That's a change management problem, not a technology problem. Which is, I suppose, always the way with technology in hotels.

Frequently Asked Questions

How long does it typically take to build the data infrastructure needed for hotel AI applications?

Marcus Devereaux's experience, eighteen months to build a customer data platform integrating PMS, CRS, POS, and loyalty systems before deploying a single AI use case, is broadly representative of what mid-size hotel groups should expect. The timeline varies based on the number of legacy systems involved and the state of existing data governance, but skipping this foundation phase is the most common reason AI projects fail to deliver their promised returns.

What is the best first AI use case for a hotel group to pursue?

Revenue management is consistently the most recommended starting point because the return on investment is direct and measurable, making it the easiest case to build internally. A successful revenue management AI deployment, demonstrable RevPAR improvement against the competitive set, generates the organisational credibility needed to pursue more complex applications in guest personalisation, operational efficiency, and predictive maintenance.

How should hotels handle staff resistance to AI-powered decision-making?

Transparency about how the AI models work, what data they use, and where their limitations lie is essential for building the trust that makes AI adoption sustainable. The most effective implementations frame AI as a tool that handles high-frequency routine decisions, freeing experienced staff to focus on strategic and contextual judgement calls. When AI recommendations are wrong, treating those moments as learning opportunities rather than failures reinforces the human-AI collaborative model rather than undermining confidence in the technology.