“Surely that can happen automatically” - the sentence comes up in almost every business. Usually when someone transfers the same data from one program to another for the third time that day. The wish is justified. But automation is not an end in itself. A badly chosen workflow does not get better when automated, only faster at being wrong.

This guide helps you sort things out: which workflows are suited, where the limits lie, which tools there are and what to watch.

Which workflows are suited

Good candidates usually have several of these traits:

  • They repeat. The same workflow runs many times, every day or every week, always to the same pattern.
  • They follow fixed rules. The decision rules can be written down: if the amount is below a limit, the team lead approves; otherwise management does.
  • A lot is re-keyed. The information already exists digitally and a person transfers it into a second system. More on this in the guide Stopping double data entry.
  • Data moves between programs. Shop, inventory, accounts, shipping: at every handover a connection can replace manual work.
  • Someone needs to be told. An order is approved, a delivery is late, a deadline is coming up. Notifications are often the simplest place to start.
  • Documents follow a pattern. Delivery notes, order confirmations, invoices that are always built the same way.

The more traits apply, the more likely the effort pays off.

Where the limits lie

Not every workflow that takes time is a good candidate:

  • Many exceptions. If every other case runs differently, the automation has to know every exception - or the work ends up back with a person. Then it often pays to simplify the workflow first.
  • Unclear rules. If nobody can say by which criteria something is decided, no program can either. Writing it down is then the real gain.
  • Poor master data. Duplicate customers, old items, wrong addresses: automation only spreads these errors faster. How to clean up is in the guide Master data in one place.
  • Rare workflows. What happens once a year is usually done faster by hand than set up and maintained.

Which tools there are

There is no single tool. Which one fits depends on where the workflow runs.

Interfaces between systems. Two programs exchange data directly, for example shop and inventory or ERP and accounts. This is the right route when the same exchange has to run permanently and reliably. What an interface is exactly is explained in the guide What is an interface.

Automation toolkits. Tools such as Zapier, Make or Power Automate connect common services in a few clicks: new email with an attachment, file it, notify a colleague. For small workflows between well-known services they are often exactly right. With many special rules, sensitive data or workflows the business depends on, they reach their limits. The guide Low-code: what it can do describes where.

An application of your own. When the workflow does not sit between two programs but entirely beside them - in lists, mailboxes and notes - an application helps that maps it as a whole and is connected to your systems where possible.

AI for unstructured input. Classic automation needs fixed fields. When information arrives as free text, PDF or scan, AI can read and pre-sort it - for example with incoming invoices. How far AI should act on its own is described in the guide AI agents in the business.

What to watch

An automation that runs is only half the work. What matters is what happens when it does not.

  • Error handling. What happens if a record does not fit, a program cannot be reached or a field is missing? Good automations stop instead of passing on something wrong.
  • Someone has to notice. If a case gets stuck, a named person should hear about it - not the customer asking after their delivery.
  • Documentation. Write down what happens automatically, by which rules and who set it up. Otherwise nobody will understand in a year why an email goes out.
  • Permissions. An automation acts with the permissions of an account. Give it only what it needs, and not the managing director’s account.
  • Responsibility. Every automation needs someone who feels responsible when rules or programs change.

How to start

Start with one workflow, not all of them.

  1. Collect. Ask the team where most is re-keyed, forwarded or chased up. The answers come quickly.
  2. Pick one workflow. Ideally one that happens often, has clear rules and where an error does not stop the business straight away.
  3. Write down how it runs today. With all exceptions. This often already shows that the workflow can be simplified.
  4. Choose the right route. Interface, toolkit, application of your own or AI - depending on where the data lives.
  5. Observe. After a while, ask the people involved: what has become easier, where does it still snag, was anything missed? These answers say more than any count.

In eight questions, the system check assesses whether the manual work in your business arises between programs or in workflows no program knows.