The prep list is the best first AI project

Everyone wants to start with the expensive thing. Food cost across the group. Labor optimisation. Something with a dashboard.

Start with the prep list instead.

Not because it is worth the most. Because it is the only candidate where a wrong answer costs you a case of romaine and a conversation, rather than a payroll run or a period close.

Why first projects fail

The first thing a team automates decides whether they ever automate a second thing.

Pick something big and the failure is expensive and public, and the story afterwards is "we tried that." Pick something invisible and nobody notices it working, so nobody believes it did anything.

The prep list sits in the gap. It is done by hand every single day, the person doing it can tell you within a shift whether the answer was any good, and if it is wrong you find out the same afternoon and throw away almost nothing.

That is the whole selection criterion for a first project: fast feedback, cheap failure, visible to the person who has to trust it.

What the prep list actually is

A guess, made by one person, out of memory.

Somebody looks at the walk-in, thinks about what day it is, remembers that last Tuesday was busier than it should have been, and writes numbers on a sheet. Then they are the only person who knows why those numbers.

The knowledge is real. Your prep cook knows things a forecast does not: that the school across the road has a half day, that the two-top section runs slower in August, that the chicken thaws differently when it is 112 outside.

The problem is not that the guess is bad. It is that it lives in one head, it takes twenty minutes every morning, and when that person is off, the guess is worse and nobody can say why.

The version that works

Give it three things it can actually check.

Last year's same weekday, from the POS. Not last week. Same weekday, same rough part of the season, because a Tuesday in August is a different animal from a Tuesday in November and your sales history already knows that.

What sold, by item, not by dollar. Prep is counted in portions. A revenue number cannot tell you how many pounds of anything to pull.

What you threw away. If you record waste at all, that is the correction on the guess, and it is the number nobody feeds back into next week.

Out of those three you get a starting list. Not a decision. A sheet somebody argues with over coffee instead of a blank one they fill from memory.

Do this on one station

Not the whole kitchen. One station, two weeks.

Pick the one with the most waste or the most frequent 86s, because those are the two symptoms of a guess going wrong in either direction.

Print the suggested numbers next to the ones your prep cook writes. Say nothing. Let them differ for two weeks.

Then sit down and go through the days where the gap was biggest. That conversation is the entire project, and it is worth more than the sheet: it is the first time anyone has written down why the number is what it is.

What you will find

Two things, most of the time.

The suggestion is close on ordinary days and badly wrong on the odd ones. That is fine and it is the point — the odd ones are where your cook's knowledge is, and now you know which days those are and what makes them odd.

And the reasons your cook gives are not in any system you own. Half day at the school. The delivery that lands at eleven. The section that turns twice on Thursdays. Write those down. That list is the actual asset, and you keep it when you change POS.

Who still owns it

The prep cook writes the prep list. That does not move, and any tool that tries to move it will be confidently wrong on exactly the day it matters.

What changes is that they start from arithmetic instead of memory, and that the arithmetic is written down where the next person can read it.

---

Not sure prep is where your week is going? Take the free check — eight questions, no email, and it names the two things worth looking at first in your building.