HOTELS Magazine ran a roundtable last month with executives from Davidson, Remington, CoralTree, Stonebridge, and Outrigger — some of the sharpest operators in the business — and the headline coming out of it was this: AI won't fix bad data. You have to fix the data first.
I read it and thought: yes. And also: most hotels haven't fixed the data problem. We've been talking about fixing it for years. And AI is about to make the gap between companies that have clean data and companies that don't much, much wider.
Matthew Bright from Davidson Hospitality said it plainly: "You can't just throw AI on top of data and expect it to come back with some kind of revolutionary information." Tim Thilleman from MIS Computer Corp was even more direct: "If they can't do Power BI, how are they going to do AI?"
These are smart, experienced operators. They've seen enough technology implementations go sideways to be appropriately skeptical. And what they're pointing at — the actual problem, underneath the AI conversation — is something I deal with every day: hotel data lives in silos, systems don't talk to each other cleanly, and the people who need the information either can't get to it or don't trust it when they do.
What the data problem actually looks like
James Wilson from Outrigger described it well: "There's a lot of siloed work occurring. Data sits in different buckets that don't correlate across departments."
That's polite language for what I'd describe more bluntly: your PMS, your POS, your GL, your labor system, and your procurement system are all keeping their own records — and they don't agree with each other. Not because anyone did anything wrong. Because they were built by different vendors, at different times, with different data models, and nobody ever took the time to map how information moves between them.
So you end up with: resort fees that post correctly in Opera but land in the wrong GL bucket in ProfitSword. F&B revenue that closes in Toast at 11pm but hits the GL with a one-day lag that makes every daily report slightly wrong. Labor costs that are allocated by department in your HR system but flow into finance as a lump sum because nobody configured the cost center mapping.
None of these are catastrophic individually. But they compound. And when an executive asks a question — "Why is housekeeping labor $40K over this month?" — the answer requires manually reconciling four different systems before you can even start answering the actual question. That's not a reporting problem. That's a data foundation problem.
The AI moment we're actually in
Here's what I think is going to happen over the next three to five years.
The companies that already have clean, connected data — the Davidsons and the Remingtons of the world, the ones with real IT infrastructure and dedicated BI resources — are going to use AI to do things that genuinely change how they operate. Remington's Nick Clark described dropping a CSV of guest reviews into an AI tool alongside service call logs, out-of-order rooms, and facilities data to understand why guest satisfaction changed. That's real. That's useful. That's the kind of analysis that used to take a week and now takes an afternoon.
The companies without that foundation? They're going to spend money on AI tools that surface garbage or confusing outputs, conclude that "AI doesn't really work for hotels," and then wait for the next cycle. Dana Cariss from CoralTree said it honestly: "I genuinely don't think we will move quickly enough as management groups or ownership groups."
He's not wrong — but the constraint isn't time or money. It's that the foundation work is unglamorous. Nobody wants to spend three months fixing PMS-to-GL mapping and reconfiguring ProfitSword cost centers. It doesn't make a good slide for an ownership presentation. But it's the work that makes every downstream tool — AI or otherwise — actually useful.
Where this lands for me personally
I'm going to be direct about something: the work described in this roundtable is the work I do.
Not the AI part. The foundation part. The part that has to happen before AI can help.
When I work with a management company or an ownership group, the first question is always the same: do the systems agree? Not "do you have good reporting" — that's downstream. Do the systems agree? Does your PMS revenue match your GL? Does your labor system reflect the same hours your GM is looking at? Does your POS settlement reconcile to your F&B revenue line without a manual adjustment every month?
Usually the answer is: close, but not quite. And the "not quite" is where all the manual work lives. The hours of reconciliation at month-end. The Excel bridges that only one person knows how to run. The ownership report that required three days to build because the data didn't come out clean.
Fixing that is not exciting work. But it's leverage work. Fix the data foundation and everything downstream — reporting, forecasting, and yes, eventually AI — gets easier.
The question worth asking right now
If you're a finance leader at a management company or ownership group, here's the question I'd put to yourself before you start evaluating any AI tools:
Can you answer the question "Why was rooms labor $30K over budget last month?" in under two hours, using data your team trusts, without a manual reconciliation? If yes, you probably have a foundation that can support better tools. If no — if the answer requires pulling from three systems, a phone call to the GM, and a manual check of the schedule — you have a data problem that no AI tool is going to fix.
That's where to start. Not with the AI conversation. With the data conversation you've probably been postponing.
One more thing from the roundtable
Nick Clark said something near the end that I think is the most important thing in the entire piece. He was asked what AI had replaced for him, and his answer was: "Reports, phone calls, trying to go through things, getting prepared for presentations."
That's the promise. Not replacing people — replacing the administrative layer that keeps people from doing the actual work. For a finance person, that means less time reconciling and more time analyzing. Less time building the bridge and more time explaining what it means. Less time hunting for why the numbers are off and more time acting on what they're telling you.
But you can only get there if the data underneath is clean. Which means the most valuable thing a finance leader can do right now is not evaluate AI tools. It's ask an honest question about whether the foundation is actually there.
Most of the time, it isn't. And that's a very solvable problem — just not a glamorous one.
If this resonates: The data foundation work — PMS-GL mapping, labor system configuration, close process cleanup — is exactly what the Integrations and Systems engagements at RoomToProfit cover. If your team is living in the manual reconciliation layer, let's talk about what clean looks like.