Essay No. 05
In late August I was invited onto a panel at me&u’s Unwrapped event, the last stop on their national tour, in a room full of people who run venues for a living.
Before the panel started, one of the company’s co-founders, Adrian Osman, got up and showed the room what they’d built. Reservations through to voice AI, live product, about to land in venues. It was good. The kind of guest-facing system that makes a booking feel effortless from the customer’s side, which is a much harder thing to build than it looks.
Watching it, I kept thinking about a different problem. Not the one on stage.
Everything me&u demoed that night solves the guest-facing half. The booking, the ordering, the voice on the phone saying yes we have a table. What it doesn’t touch, and isn’t trying to, is what happens in the office after the guest has gone home. The roster. The supplier invoice. The report nobody’s had time to read properly since March. That’s a different layer, and it’s the one most operators are still doing entirely by hand.
Which is closer to the question I actually got asked on that panel: what’s the real blocker stopping operators from using AI properly? Not the guest-facing stuff, the back-office stuff. Everyone wants to answer that question with time, budget, or trust. None of those are the real answer.
I’ve sat across from enough operators now to know what it is. Most of them have never once sat down and written out where their week goes. Not in general terms. Line by line. The invoices. The roster juggling. The reviews nobody’s replied to.
You can’t point a tool at a problem you haven’t named. That’s not a catchy line, it’s the entire bottleneck. AI is extraordinarily good at doing a task you can describe. But you can’t automate a feeling. And a feeling that something’s off, without the words for what, is exactly where most operators are standing.
So half of what I do with a new client isn’t AI at all. It’s the first conversation. Dragging the real problem into the light, out of the vague shape it’s been living in, into something specific enough to hand to anything. A person, a system, doesn’t matter. Once it’s named properly, the AI part is almost an afterthought. Easy, even.
That’s the bit nobody sells you, because it doesn’t look like a product. There’s no dashboard for it. It’s just a conversation where someone finally says the actual sentence out loud, instead of the vague version they’ve been carrying around.
Here’s the tool I’d tell every operator to steal, and it costs nothing.
Not a hospitality product. Not anything built for venues, not what was on stage that night. A plain chat model, used the way you’d use a sharp new manager you trust enough to think out loud in front of. You describe the problem the way you actually experience it, messy and half-formed, and it asks you the next question. The one you’ve been avoiding.
The value is in the questions it makes you sit with, more than anything it answers.
I use it that way before I build anything, every time. Not to generate a solution. To interrogate my own description of the problem until it’s sharp enough to be wrong about, which is different from being too vague to be wrong about at all. A vague problem can absorb any solution and still not get fixed. A named one can finally fail cleanly, which means it can finally get fixed properly.
A vague problem can absorb any solution and still not get fixed.
Most operators skip straight past this part. Understandably. It doesn’t feel like progress. Sitting with a question instead of installing a tool feels like standing still while everyone else is moving. But the ones who get real value out of AI were never the technical ones. They’re the ones who can describe their own business clearly. Everything downstream of that is comparatively easy.
I think about the operators who come to me already knowing exactly what they want automated. It’s rare, and when it happens the build is fast, almost boring. Then there’s everyone else, who arrives certain something needs to change and unable to say what. That’s not a lesser starting point. It’s the honest one. Most people are in the second group. I was, for years, before I had the language for any of it.
The tool doesn’t care which group you’re in. It’ll happily automate a badly named problem just as fast as a well named one, and hand you back something that solves nothing, dressed up as progress. That’s worse than doing nothing, because now you’ve spent the afternoon and you still don’t know what was actually wrong.
What me&u put on that stage was proof the guest-facing side is being solved properly, by people who are good at it. The other half has nobody working on it with the same seriousness, and I think that’s because it doesn’t demo. There is no moment on a stage where a supplier invoice reconciles itself and someone films it on their phone.
There is no moment on a stage where a supplier invoice reconciles itself and someone films it on their phone.
The back office is a different job. Name it first. Everything else is the easy part.