Hospitality Is Not a Data Problem
Where AI belongs in a food business, and the one place it must never go.
I have watched AI production tools pitched to operators. Recipe generation, menu development, the machine that will design your next dish from your sales data. I thought they were wrong at the time and I have not changed my mind, though I want to be precise about why, because the obvious objection is the weak one.
The obvious objection is that the food would not be good. That is a capability claim, and capability claims age badly. Models improve. Anyone whose argument rests on the machine can’t do it yet is writing a sentence with an expiry date.
The argument is not that the machine cannot do it. The argument is that the world has competition.
Three things, not one
A great food brand combines the art of making the perfect product with the science of doing it consistently, backed by SOPs, and a clear discipline around margin.
Everyone treats those as one system. They are not. AI stands in a completely different relationship to each of the three, and almost every mistake in this sector comes from failing to separate them.
Art. The recipe. The menu architecture. The judgment that this plate leaves the pass and that one does not. This is taste, in the literal and the structural sense.
Science. The SOP. The prep sheet, the temperature, the sequence, the checklist. Every one of these encodes a judgment that a human being already made and then wrote down.
Margin. Procurement timing. Cover forecasting. Wastage. Roster hours against expected volume. Arithmetic under uncertainty.
Three inputs. One product. Entirely different exposure to automation.
Where the machine belongs
Margin is the machine’s home and it should live there permanently.
Forecasting covers, timing purchase orders, optimising a roster against demand that has not happened yet — these are problems of arithmetic under uncertainty, they are dull, they are enormous in aggregate, and no human being does them well. A restaurant group that runs its procurement on instinct is burning money in a way that is invisible on any single day and ruinous across a year.
Give the machine every one of these. Give it all of them.
Where it can help, and where it stops
An SOP is different, and the difference is subtle enough that it is worth being exact.
A machine can enforce an SOP beautifully. It can check the temperature was logged, the checklist was completed, the sequence was followed. It can do this across two hundred outlets at once, which no regional manager can.
A machine cannot write one. An SOP is the fossilised record of a decision someone made about what good looks like — how long, how hot, in what order, why this order and not the other one. The authoring is the entire value. The enforcement is the commodity.
Automate the enforcement. Never confuse it with the authoring.
Why production is the line
Now the part that matters.
Every input to a food business is purchasable. The SOP can be reverse-engineered by anyone who eats the food twice. The supply chain is available to your competitor at a similar price. The software you bought is on sale to the person across the street. The site, the fit-out, the aggregator contract, the labour — all of it is a cheque.
The recipe is not. The hand that developed it is not. The accumulated judgment of the person who tasted the thing four hundred times and knew, on the four hundred and first, that it was finished — that is not for sale, and it is the only input in the entire business that is not for sale.
Which is why the capability objection is backwards. The stronger the model gets at recipe development, the worse this becomes, not the better. A machine that generates a good dish generates it for everyone. You have not acquired an advantage. You have retired one — traded the only non-copyable asset you owned for an efficiency gain that your competitor purchases on the same terms, from the same vendor, next Tuesday.
You cannot build a moat out of something that ships as a subscription.
The test
So the line is not front of house against back of house. That is an org-chart answer to a question that is not about org charts, and it gives the wrong result in both directions — it hands the kitchen to the machine and reserves judgment for the people who greet you at the door.
The test is simpler. Does the task have taste in it.
Taste is distributed across the whole building. It is in the recipe and it is in the decision to comp a table. It is not in the purchase order. Wherever taste is present, a human decides. Wherever it is absent, the machine should run unattended and unsupervised, and you should stop paying someone to do arithmetic.
The same rule, applied to me
I run a company that sells AI-executed growth, so I should say plainly that this rule does not exempt me.
Growth has taste in it. Positioning has taste in it. The decision about which thousand people receive a message, and what that message concedes, and what it refuses to claim — all taste, all judgment, none of it available from a model.
What AI has done is make capability cheap and judgment scarce. Anyone can generate a thousand emails now. Almost nobody can decide which thousand people should receive them. The bottleneck did not disappear. It moved upstream, and it moved to the exact place where nobody has automated anything.
So the structure that survives is humans deciding and machines executing. Not humans reviewing machine output, which is the arrangement almost everyone has built, and which is a different thing wearing the same words. In that arrangement the machine has already made the decision and the human is auditing it, tired, at volume, at the end of a queue. Direction is not a detail. Direction is the whole argument.
What this is not
None of this is scepticism about the technology. I have built a business on it.
It is that a technology which lowers the cost per unit of output has very little to offer a business whose product is the unit — and quite a lot to offer the ninety per cent of that business which is not the product at all.
Put the machine in the ninety per cent. Leave the ten alone. The ten is why anyone comes back.