Where business process automation actually starts in an SME

Every automation project that fails in a small company fails the same way: somebody bought a tool before anybody wrote down the process. January is a good month to do it in the right order.

Start by watching, not by asking

Sit next to the person who handles incoming supplier invoices and watch them do twenty. Do not interview them — watch. People describe the process they think they follow, which is never the process they actually follow.

You will see things nobody would have told you. That three suppliers send PDFs which have to be retyped because the numbers are in an image. That one customer’s reference format breaks the import every time. That a colleague keeps a private spreadsheet because the official system cannot do something.

That hour is the whole foundation. Everything after it is easier.

Then score the candidates honestly

A task is worth automating when four things are true at once:

• It repeats. Weekly at minimum, ideally daily. A quarterly task is almost never worth the setup.
• The rules are stable. If the answer depends on who is asking and how they feel about it, you are automating a judgement, not a process.
• A mistake is visible and cheap. Automate where an error shows up quickly and can be corrected. Not where it silently enters the accounts.
• Someone currently hates doing it. Underrated. You need an internal ally, and the person who dreads the task on Monday morning is that ally.

Then, and only then, the tool question

By this point the tool question is usually small. Most SME processes need one of three things: something to read documents, something to move data between two systems, or something to draft text a person then approves.

If you cannot say which of those three your first project needs, you are not ready to choose a vendor, and every demo will look equally convincing.

The rule of one

One process. Finish it. Let it run for a month before starting the second.

Companies that automate three things at once finish none of them, because the hard part is never the technology — it is the exceptions, and exceptions only surface once something is actually running.