Role
Strategist · Product Lead · AI Designer
A living operational memory for boutique hotels. Built from the signals staff already produce — every shift note, every observation, every decision — structured into intelligence that compounds over time.
I started with a question: why do well-run hotels keep making the same operational mistakes?
The answer wasn't a lack of data. It was that the data was never connected to decisions — and decisions were never connected to outcomes.
When I started researching how boutique hotels actually operate, I kept finding the same pattern. A front desk manager notices a guest seems uncomfortable but has no way to log it. A bartender remembers that a certain type of guest always orders the same thing — but that observation never reaches anyone.
The problem wasn't that hotels lacked data. It was that the most valuable operational knowledge — the human kind — had no place to live. I designed Nexus to capture exactly that, before it disappears.
Every recommendation comes with a predicted outcome — and a record of whether it held.
Every observation, every decision, every outcome stays in the system. The knowledge compounds.
The people who run the hotel train the system — not just the data their tools produce.
When staff rotate — and they always do — the hotel starts from zero.
Nobody knows what was predicted, what was tried, or whether it worked. The same mistakes repeat.
Data exists everywhere — PMS, POS, staff observations — but nothing connects it into something a GM can act on tonight.
Two principles, set early in the design. They're what allow an operational AI tool to sit inside a hospitality business without becoming a surveillance system or a confidence trick.
The biggest objection I heard from GMs wasn't about the technology — it was about trust. They needed to know the system couldn't be used against their guests. So Nexus never stores who did what. Identity is stripped at the moment of capture. What remains is a behavioural pattern, not a person.
Early in the research, I noticed that GMs had learned to distrust recommendations — because nobody ever proved they worked. Every suggestion Nexus makes is a bet: the prediction is recorded before the outcome is known, then the result is written next to it automatically.
Nexus reads what is already there, stamps each event with zone, time, and category, and strips the identity at the edge — before it travels anywhere.
The first structured pilots are running with hotel operators in the Netherlands, Switzerland and the UK. The system is in active training — learning from real operational inputs, refining its models with every shift.
This is not a prototype being tested for feasibility. The question being answered now is whether the system learns fast enough to be useful within a single season.