Right now, there’s only one AI ingredient under your control.
Let me explain.
Every AI system is made of three things.
- Data
- Model
- And compute.
Most people aren’t going to make their own chips, or train a frontier model. Instead, they’ll rent them from enormous companies like everybody else. That means no control over your compute or your model.
That leaves data.
Data is the only ingredient you control. And it’s the only place a lasting advantage can come from.
But most hospitality brands are doing very little with. And if you want to get the most out of the newest AI tools, you’re going to have to fix that.
More data in, more value out
An agent is only as useful as what it knows.
Ask a general-purpose model with no extra context to help a general manager in Leeds handle a return-to-work conversation and you'll get something generic and vaguely American.
Give the same model your absence policy, that person's history, their manager's notes, the site's cover arrangements, and last winter's pattern of short-term absence across the estate — and you get something actually useful.
All things being equal, the more of your data you can show the model, the more value you get back.
That's it.
The quality of your data sets a ceiling on what AI can ever do for you.
So the strategic question isn't "which AI tool should we buy?" It's "what data do we have, and will we be able to use it?"
And when you ask that, you’ll see two nasty surprises are waiting.
Surprise one: Do you own the data you think you own?
Ask most hospitality leaders who owns their data and they'll say: we do, obviously.
Then look at what happened with distribution.
Book a room through Booking.com or Expedia and the hotel doesn't get the guest's email address.
It gets a proxy — a relay address that sits between the two of them.
Expedia's version expires shortly after checkout. So an operator pays a double-digit commission on the booking, hosts the human being for three nights, cooks their breakfast, and still can't email them afterwards to ask them back.
The relationship is theirs. The data isn't.
It's not just hotels – it’s other sectors too.
Deliveries went the same way.
Every order through an aggregator belongs to the aggregator. Not the name, not the email, not the order history. You cook the food. They keep the customer.
It’s not some kind of conspiracy. It’s just the contract, signed by people who were focused on the commission rate and didn't argue about the data clause, because in 2016 the data clause looked like boilerplate.
It wasn't boilerplate. It was the most valuable term in the agreement.
Now go and look at your People tech contracts and ask the same question, properly this time. Not "do we own our data" but:
- Where does the contract say the data lives, and who has the right to use it?
- Can we export all of it, in a usable format, whenever we want — or only the reports the vendor chooses to give us?
- What happens on the day we leave? What do we get back, in what format, and what does it cost?
- Is our data helping to train something that also serves our competitors? And are we getting anything in return?
People Directors who actually go and read the clause are frequently unpleasantly surprised.
Surprise two: Owning it isn't the same as using it
Say the contract is fine and the data is legally yours. You still have a problem.
Hospitality has spent 20 years building connectivity standards — OpenTravel, HTNG, HEDNA.
There's an active industry conversation about the fact that they overlap and duplicate each other. But notice what all of them are for: distribution and the guest. Rooms, rates, availability, bookings.
There is no equivalent effort for employee data. Nothing close.
So your people data sits in pieces. Some in payroll. Some in the rota. Some in the applicant tracking system nobody likes. Some in a spreadsheet on a regional manager's laptop. Each one with a different definition of a leaver, a different definition of a site, and its own opinion about what counts as full-time.
Which is why the sector can't answer a simple question about itself. Look at published turnover figures for UK hospitality. One 2024 analysis puts it at 38.7%. The Caterer has reported figures implying about 6% of people leave every month. A study of more than 35,000 hospitality employees put annual turnover at 67% in late 2025, down from 75%.
Those aren't rounding errors. That's an industry that cannot agree on its single most important people number. Not because anyone's lying — because the data is fragmented, defined differently everywhere, and nobody can reconcile it.
You can't build an advantage on top of that. And no agent will save you from it.
The bigger problem: you're not generating the data at all
Here's the part almost nobody is talking about.
Even if you fixed the ownership, fixed the integration, and got every system talking — you still wouldn't have the data that matters most.
Think about what you'd need to know to give someone a genuinely personal experience at work. Not a mail-merged first name. Personal.
You'd need to know that they're vegan. That they're dyslexic, and that a wall of dense text is why their compliance training keeps timing out. That they want to be a head chef by 30 and nobody has ever asked them. That they've got a four-year-old and a father with dementia, and that Tuesday mornings will make or break their next 12 months. That they coach a junior football team on Sundays and would be brilliant at inducting new team members, if anyone had thought to connect those two facts.
None of that is in your People tech stack. Not one field of it.
It isn't there because legacy systems were built to answer the employer's questions — who's employed, on what terms, are we compliant — not the employee's. Consistency, compliance, and control. A system of record for the contract, not for the person.
And you can't personalise an experience using data you never collected.
This is exactly why we built the Youda HRIS. Not as another record of contracts and cost codes — there are plenty of those, and some of them are decent. We built it to be flexible enough to capture, hold, and analyse the things that actually matter to a person: their circumstances, their ambitions, their preferences, the shape of their life outside work. The stuff that decides whether they stay.
That design choice comes with an obligation. The system has to be safe, honest about what's held and why, and useful to the person giving you the information — or they won't tell you any of it, and quite right too. Personalisation isn't something you extract. It's something people opt into because they get something back.
And then the thing that changes everything
Now the ambitious part.
Personal data tells you who someone is. Behavioural data tells you what they do — and, if you capture it with enough context, why they did it.
When agents do the work, the record is a byproduct. An onboarding agent knows exactly which step lost people and how long each one took. An absence agent knows what happened in the weeks before someone stopped coming in. A career conversation agent knows what someone said they wanted, when they said it, whether anything followed, and what they did six months later. Not a survey score six weeks after the fact. What actually happened, as it happened, joined to what happened next.
Do that across an estate, over years, and you have something this industry has never had: the raw material for a causal inference model.
That's what Youda is building towards, and we'll be straight about where we are. The model needs data, and the data needs time. There's no shortcut, and we're not going to pretend there is.
But think about what it's worth on the other side.
Imagine knowing exactly what worked. Who it worked for. When it worked. And what it was worth to the business, in money.
That's the end of hit and hope.
You'd stop defending your budget as a cost and start allocating it against expected value, like every other function that gets taken seriously. You could forecast. You could walk into an exec meeting and say: this, at these sites, for these people, returns this — and here's the evidence. You'd stop relying on being liked and start relying on being right.
That's the difference between a People Director who administers and a People Director who decides.
The Time Is Now
Here's why this is urgent rather than merely interesting.
A data advantage is time-based, and time-based advantages are the only kind that hold. A competitor can copy your benefits package in a quarter and your employer brand in a year.
They cannot copy three years of behavioural history about a workforce they don't employ. They can't buy it, borrow it, or backfill it. They can only start their own clock, three years behind yours.
Which means the operators who start generating this data in 2026 will be structurally ahead of the ones who start in 2029, and that gap will widen rather than close.
The uncomfortable version: whoever holds your data in 2029 will hold a lot of power over your business. Right now, you still get to decide who that is, and on what terms.
Go and read the clause.
If you're building a People function for a business with a large frontline team, book a chat with us. We can help you with what data you're generating, what you actually own, and what you could do with it if you could see all of it.