What to automate in a small factory, and what to leave alone

When someone says "automate the factory", almost everyone pictures a robotic arm. That is what shows up in the videos.

But in a factory of twenty people, the money lost to a lack of automation is almost never on the shop floor. It is in the office next door, in a number somebody is typing for the third time.

Automation is not robots

The word drags in assembly-line imagery, and that derails the conversation in the first minute. In practice, automating means getting a piece of data to travel on its own from where it originates to where it is needed.

Think about an order. It arrives by message. Someone copies it into a spreadsheet. The production order comes out of that. The delivery note comes out of the production order. The invoice comes out of the delivery note. That is the same data five times, and every copy is a chance to get it wrong.

The difference against machinery is enormous in both cost and risk. Automating a machine means capital, months of lead time and stopping the line. Automating the flow of information means software, weeks, and stopping nothing.

What is worth automating

Any data typed in twice. This is the most profitable rule there is. If a number lives in two places and a person moves it by hand from one to the other, that is where the hours go and that is where the errors come from.

The Monday report. Anything a person assembles by hand on a fixed schedule. If it can be done from a template that never changes, it can be done without you.

Stock alerts. You do not need a system that buys for you. One that warns you before a shortage becomes urgent is enough — urgent is the expensive part.

Documents that always come out the same. Delivery notes, invoices, certificates. The data changes; the shape does not.

The traceability your regulations demand. Batch, date, shift, raw material. If you are required to record it, recording it by hand means paying twice: once in time, once in the panic on audit day.

What to leave alone

A process that changes every month. Automating a moving target means rebuilding the automation every month. Stabilise the process first, automate it second.

Anything where the exception is the rule. If four orders in ten are a special case, you are not going to automate a process — you are going to write forty rules and maintain them forever.

Decisions that come down to judgement. Accepting or rejecting a borderline batch, giving a date to a customer who is already running late, choosing what to make first when everything is urgent. A system can hand you the numbers to decide with. Not the decision.

Anything done once a quarter. Four times a year, twenty minutes each, is eighty minutes. Automating it costs more than doing it.

The person who also catches the mistakes. This is the expensive one, and nobody sees it coming. Whoever types the same figure three times is also the person who notices when something does not add up. Automate the task without putting a check in its place and you gain speed while losing a quality control that was never written into any procedure — but was real.

The five-minute calculation

Three numbers: how often it happens, how long it takes, and what it costs when it goes wrong.

A ten-minute task, five times a week, is forty-three hours a year. More than a full week of one person's time, spent on something nobody has written down as a problem because it never takes more than ten minutes.

If a mistake there also ships the wrong pallet or stops a line, add that separately.

Rule of thumb: if it does not reach twenty hours a year and has never caused an expensive problem, leave it alone. There are better things to automate first.

Where to start

Wherever the data is typed in twice. Every time.

It is the easiest to measure, the easiest to undo if it goes badly, and the least likely to start an argument inside the team — because nobody defends typing the same thing twice. And there is something that matters more than it sounds: the first automation that works buys you permission for the second one. Start with the hard one, land it badly, and there will not be a second.

The mistake that costs the most

Automating a broken process does not fix it. It makes it fail faster, at higher volume, with fewer people watching.

The test is simple: before automating a step, you have to be able to describe it in one sentence, with no "except when". If you cannot, the problem is not a software problem, and no tool is going to solve it.