What's worth automating in a business - and what's better left alone

Language models have changed what can be automated. Before, a machine could only handle data in a fixed format - spreadsheets, forms, exports. Today it reads a freely written email, sorts it and drafts a reply. So almost everything can be automated. The question is no longer whether it can be done, but whether it pays off.

Test one: does it repeat?

Automation pays off on work that happens over and over in the same way. Sorting incoming enquiries, replying to reviews, copying invoices into the accounting system, a weekly report from several systems. A task that comes up twice a year and looks different every time is cheaper to leave to a person - the automation wouldn't have time to pay for itself.

Test two: does it have a clear input and output?

A good candidate can be described in one sentence: this comes in, this should come out. An email with an enquiry comes in, a row in a spreadsheet and a draft reply should come out. An invoice in PDF comes in, a line in the accounting system should come out. When neither the input nor the output can be named, there is usually no settled process - and that needs to be clarified before anything is built. Automation doesn't organize chaos, it only speeds it up.

Test three: how many hours does it cost today?

This is where most companies underestimate themselves, because manual work is spread out. Two people who each spend six hours a week sorting emails and retyping orders add up to 576 hours a year - over two and a half months of one person's full-time work. Nobody sees it as one big problem, because it happens in small pieces. Before we start on anything, we work out this number. The calculator on the Automation page gives a rough estimate in a minute.

What not to automate

Decisions that someone is accountable for and where a mistake is expensive: custom price quotes, responses to a serious complaint, anything with legal consequences. Here automation can prepare the materials, but a person should make the decision. The second group is processes that are changing right now - automating something that will look different in three months means paying twice. And the third: things where personal contact is valuable in itself. A regular customer who calls about an order may not want to talk to a machine.

Where to start

The best first automation is a boring one. Not the most ambitious, but the one that passes all three tests and annoys someone the most today. It pays for itself quickly, the team gets used to it, and it shows how automation works in practice in your business. You can then build on it. With us, every deployment starts with a free process review: we go through what you do by hand today, count the hours and tell you which of it is worth taking over - and which is not.