Your spreadsheet is a model. It just doesn’t know it yet.
Food companies tell us their data is a mess. Then they open a spreadsheet with six years of history, colours that mean something and formulas nobody touches. A note on what is already in it.
JordanOsakaAlmost every food company we meet says some version of the same sentence: “We’re not ready for AI. Our data is a mess.”
Then someone opens a spreadsheet. It has six years of history, a tab for every season, colours that mean something, and formulas nobody dares touch. It turns out the data isn’t a mess. It has just been living somewhere nobody calls a database.
This post is about that spreadsheet: what is already in it, what reading it properly turns up, and what changes once it knows what it is.
Meet Kees
Kees runs the stores at Van Dijk Fruit, an apple grower-packer in Geldermalsen, in the Betuwe. (The farm is made up, but you have met Kees.) They grow four apples: Santana, Elstar, Boskoop and Jonagold. After picking, the apples go into six cells. Three are ordinary cold cells. Three are ULO cells: sealed, low-oxygen rooms where an apple can keep for most of a year. The catch is that once a ULO cell is opened, everything inside it is on the clock.
Since 2020, Kees has kept one workbook. Every morning at six he opens it, adds what went out, and checks the temperatures. If a lot has been sitting too long he colours its row yellow. On Fridays the sheet goes to the buyers: a supermarket chain, a fruit trader, a German importer, a UK importer and the farm shop.
Ask Kees whether he has a data model and he will laugh. He has a spreadsheet.
What’s already in there
Read the workbook the way a model would, and it has already done most of the work.
The things. Each row is a lot of bins. One column names the cell, another the apple, another the buyer it is promised to. That is four kinds of thing and how they connect, which is all a data model is.
The rules. One formula says how old a lot is. Another works out its sell-by date: the picking date, plus how long that apple keeps in that kind of cell, minus a fifth if the cell runs warmer than 1.5 °C. Kees wrote that formula after a bad winter. It is still the best rule anyone on the farm has.
The rule nobody wrote down. The yellow fill. Kees colours a row when a lot should go first. That is a rule too. It just lives in his head and only runs when he is at the desk.
The rule in a comment. On one row there is a note: ULO 2 niet open vóór februari, “don’t open ULO 2 before February.” Anyone new on the farm would have to find it by hovering over the right cell.
None of this has to be typed again. It has to be read.
What reading it turns up
Reading it carefully also finds the cracks every long-lived spreadsheet has.
- Cold cell 2 is typed two ways. Three rows say 2.4 °C. One says 1.0 °C, because someone checked the compressor, decided it was fine and typed over the number. That lot’s sell-by date now skips the warm-cell penalty and lands on New Year’s Day instead of mid-December.
- A date where the formula was. One Boskoop lot has “31 March” typed into the sell-by column. The UK buyer would like them until spring. The formula says early February.
- A lot with no date. One Jonagold lot has an empty picking date. Excel counts from 1900, so the lot is 46,339 days old, and its sell-by date is May 1900.
- An old lot nobody coloured. Kees’s yellow rule works at about 60 days for every apple. Santana keeps for about 30. One Santana lot is 46 days old, past its date, and still white.
None of these are failures by the person keeping the sheet. A spreadsheet can’t tell you that two cells mean the same thing, that a number used to be a formula, or that a colour stands for a rule. It holds values and nothing about what they mean.
What changes when it knows
Turn the same rows into a model and the questions change from “give me an afternoon” to “here you go”. You can try every one of them in the apple example.
How much has to be sold before New Year? 119 bins, about 38 tonnes, and two lots are already past their date. Sorting the sheet by its sell-by column gives 32 tonnes. That misses the lot with the typed-over temperature and counts the lot from 1900.
Can we fill the supermarket’s Christmas promotion without opening a ULO cell? Yes. The order is 150 bins of Jonagold, packed the week before Christmas, and it has to stay good until mid-January. Two lots in the cold cells cover it, with 195 bins to spare. Opening ULO 2 would have put 188 bins of Elstar and 375 bins of Jonagold on the clock in December, which is exactly what Kees’s comment was there to prevent.
Which lots should go first? Four. The two Santana lots past their date, and two Elstar lots in the warm cell that have to go by mid-December. The sheet had coloured one of them.
What is the tired compressor in cold cell 2 costing us? Up to 30 days of selling time on 70 tonnes of apples, worth about €56,000. Drag the cell’s temperature back under 1.5 °C in the example and watch the dates move back into January.
Every answer lists the rows it used, so nothing is a black box. The rules are Kees’s own, kept exactly as he wrote them, including the 0.8 and the 1.5 °C.
What it isn’t
It isn’t a new system that Kees has to learn before anything works. The first version reads the workbook he already keeps, and he can keep keeping it.
It isn’t an AI that makes things up. The model works out sell-by dates the same way the sheet does, only consistently, on every row, every morning. Where AI earns its place is in reading the sheet in the first place (spotting that column C is a list of cells, that the yellow fill is a rule, that a comment holds an instruction) and in letting anyone ask a question in plain words.
And it isn’t a judgement on the spreadsheet. The spreadsheet is the reason there is anything to model at all.
Being “AI-ready” is mostly five habits
The industry talks about AI readiness as if it were something you buy. Mostly it is a handful of habits, and most growers, traders and processors already have some of them.
- One row per thing. One lot, one row. Not one row per lot, plus a summary row, plus a row for the bit that was resorted.
- Type things once. A cell’s temperature belongs to the cell. Put it in one place and look it up everywhere else.
- Keep rules in formulas, not in your head. If you colour a row by eye, write down when you do it. That sentence is a rule a model can run.
- Never type over a formula. If the buyer wants a different date, add a column for what was agreed. Don’t overwrite what the apples will actually do.
- Date everything. A lot with no date is a lot nobody can plan.
None of this needs software. It needs a morning and a bit of discipline. Do it, and the model is mostly there.
How we’d do it with you
This is the work we like best, and it starts from what you already have, not from a blank system.
- Send us the sheet. The real one, with the colours and the copies.
- We read it and show you what is in it. Usually within a week: the things, the rules, the rules nobody wrote down, and the cracks. Most teams find this part useful on its own.
- We fix the cracks together. Mostly the five habits above, on your own sheet.
- We build the first question. The one that currently costs someone an afternoon every week. For Kees it is “what has to go before New Year?”
- Then we build around it. The promise to the buyer, the alert when a cell runs warm, the weekly plan. Your team keeps working the way it does today, with fewer afternoons lost to arithmetic.
The model was already in the sheet. We help you build around it.