Clean Data and Good Governance: The Real Foundation for Using Gen AI in Hospitality

Overall, there is no difference between preparing a dataset for AI and preparing a dataset for the myriads of other tech tools and business uses. By AI here, we mean Generative (Gen) AI-next-best-word, probability & correlation-based tools. Traditional analytical tools need normalised data (third normal form etc.) using Structured Query Language (SQL) with potential ambiguities removed in the design. Gen AI is useful to analyse text documents which are, by definition, unstructured, but it still acts as a glorified "word cloud" linking items together by frequency and probability. The main hospitality systems - PMS, POS, Spa, RMS, CRS, CRM - were built on normalized databases precisely to prevent mismatches, duplication, and gaps. Even then, duplicate guest profiles remain common, eg when a frequent guest changes email or surname. The better PMSes use fuzzy matching to ease de-duplication, but this remains a governance issue.

Not all APIs are equal. Vendors often claim "open APIs" but deliver the bare minimum - eg lead guest name without full party details, or guest name without booking source. Integration should be treated as a strategy, not a patchwork: middleware, or event-driven architectures can help unify fragmented systems more effectively than point-to-point fixes. Governance is just as important-consistent guest IDs, master data management, and privacy-compliant retention rules ensure clean data at the source.

I"ve seen Gen AI tools thrown with hope and prayers at messy corporate datasets; the tools did what they could but failed on logic and timing. For instance, they didn"t understand "recognised revenue," which can only be banked when services are delivered, not sold. Cancellable rooms create revenue liabilities, and hotels still define cut-offs at night closure (2–3 am). Gen AI left to its own devices would mis-handle this. LLMs have no world model-they don"t know before/after, push/pull, or date cutovers. Agentic AI is essentially an LLM on top of other LLMs orchestrating process flows. Fine for simple cases, but brittle when exceptions arise (eg similar guest names, out-of-order rooms, midnight rollovers, missing data). Traditional RPA players like UiPath, Automation Anywhere, and Workato remain safer orchestration partners. Human-in-the-loop design-where humans review exceptions and validate AI actions-is not optional; it must be part of the architecture.

Unstructured datasets in hospitality today are ripe for Gen AI harvesting: social media posts, TripAdvisor reviews, free-text PMS notes, contact centre transcripts. These can drive sentiment analysis, guest journey mapping, operational anomaly detection, and service improvements. Structured data, however, is still best handled via BI techniques (ETL pipelines, data lakes /lake houses) with visualisation tools layered on top. Gen AI can assist here as an extra wordsmithing module, but as not the analytical engine itself.

There are no short cuts: clean, well-governed structured data plus thoughtful use of Gen AI on unstructured content is the right mix. Investment in deduplication, proper integrations, and governance pays back by reducing service errors and guest dissatisfaction, making AI the gloss on top, not the foundation.

View related World Panel viewpoint

Fergus is the former VP/Director of IT at Red Carnation Hotels (UK). Prior to Red Carnation Hotels, Boyd was VP/Director of Digital & IT in YOTEL, the innovative hotel start-up in the new “affordable luxury” sector.

Hospitality Financial and Technology Professionals (HFTP®), established in 1952, is a hospitality nonprofit association headquartered in Austin, Texas USA with offices in United Kingdom, Netherlands and Dubai. HFTP is recognized as the spokes group for the finance and technology segments of the hospitality industry with an international network of members and stakeholders.

Comments

Loading comments...