Why We Still Use Pen and Paper for Inventory, and Where AI Actually Earns Its Place

It is 2:30 in the morning. Minus twenty degrees. Me and a colleague are standing in a walk-in freezer with a tablet, trying to finish a stock count before we can go home.

Every five minutes, we have to step out. The cold kills the touchscreen. Sometimes there is no network at all, and the app just sits there, not saving anything we typed. We step back in, pick up where we left off, lose the thread again.

After an hour of this, we stop being careful. We start guessing. Put a number down, move to the next shelf, get out of the cold faster.

The next day, the inventory said we had an item we did not have.

That gap did not come from either of us being careless. It came from a device fighting us the entire time, in the one place it was always going to struggle.

Where the Tool Breaks Is Not an Accident

Cold kills batteries and touchscreens. A walk-in freezer blocks network signal by design, that is what the insulation is built to do. Try using a phone inside one.

So the one place a stock count has to happen reliably is also the one place most devices perform worst. That is not a coincidence an app can fix. It is physics.

And it is not only a cold room problem. A new team member once did a closing stock count after a ten hour shift. He had to log wastage for a tray of milk. The app’s unit field defaulted to litres. He meant to type nine hundred millilitres. He never touched the dropdown. He submitted nine hundred litres of milk wastage and moved on.

Nobody caught it that night. It surfaced weeks later, at month end, when the numbers stopped making sense and someone had to trace the whole gap back to one line, one unit, one dropdown nobody thought twice about.

Two different nights, two different mistakes, one shared cause. Put a tired person in front of a screen that asks for more attention than they have left, and something will slip. The cold makes it worse. A long shift makes it worse. A dropdown makes it worse, because it is one more decision when all someone wants to do is write a number and move on.

Pen and Paper Wins at the Point of Count

A pen does not need a battery. Paper does not need signal. You can write on it with gloves on, drop it, step on it, and it still works.

None of that is nostalgia. It is just what a freezer at 2:30 in the morning actually demands, and no app has beaten it yet.

The count itself should stay this simple. Write the number next to the item. Hand the sheet to whoever checks it. Go home.

Where people go wrong is stopping there. A simple count does not mean the system around it should stay simple too. The sheet is not the end of the process. It is the beginning of it.

Where AI Actually Earns Its Place

This is where AI genuinely helps, and where most of the industry points it at the wrong step.

At the cafe we run this at, closing stock is still written by hand on a printed sheet that mirrors the format in CafeStock. A manager takes that sheet, often photographs it, and the numbers get read and entered from there. Right now, that reading step is still mostly a person matching handwriting to item names.

The next version of that is simple. Photograph the sheet. Let a model read the handwriting. Have it fill in the same digital fields a tired employee would otherwise have had to type by hand, dropdown and all.

I want to be honest about what that step can and cannot promise. Handwriting on a sheet filled out fast, with crossed out numbers and no patience left, is not always clean. A model will get most of it right and save real time. It will also misread a number next to a smudge, or a seven that looks like a one.

So glance over it before you trust it. That is the whole rule. Not because the model is unreliable, but because anyone’s handwriting deserves a second look before a number from it goes into a report.

CafeStock still lets you type a stock count straight into digital fields, and some people will prefer that, especially for a small count done at a desk and not inside a freezer. That option staying there is not a contradiction. It just means the input method should match who is doing the inputting, and when.

This Is Not a Case Against Technology

None of this is an argument for staying away from software. It is an argument for noticing where in the process a tool gets used, and whether that spot matches the person using it.

A tired employee at hour ten of a shift and a manager reviewing numbers the next morning at a desk are not the same person. They should not get the same interface. Give the first one a pen. Give the second one everything else: OCR, reconciliation, trend reports, reorder alerts, all of it.

The mistake was never using AI in a restaurant. The mistake is pointing it at the moment someone has the least left to give, instead of the moment right after, where it can do real good with nobody’s patience on the line.

That night in the freezer, a pen would have gotten us out five minutes sooner each time. AI’s job was never to replace that pen. It is to make sure what gets written on it does not sit around for a month before anyone reads it right.