What an automation workflow is

An automation workflow is a saved set of instructions that runs by itself when something happens. When a tool says it can automate a task, it means you can describe that task as a trigger followed by steps.

What's inside one

You need three ideas. Everything else in these tools is built from them.

• A trigger. The event that starts a run. Zapier puts it this way: a trigger is an event within an app that starts a Zap.¹
• Steps. Each one does a single job, such as sending a message or looking up a record. n8n says its workflows run node by node, and can branch, merge and loop.²
• Data. Every step receives something from the one before it and passes something on.

The makers use different words. n8n calls the whole thing a workflow and each step a node. Zapier calls it a Zap. Make calls it a scenario and each step a module.²³

What "automate" means here

The workflow doesn't decide to do anything new. You decide the order of events in advance, and the tool follows it every time the trigger fires.

Zapier's own example shows the idea. A Zap starts when a new email with an attachment arrives in Gmail. It uploads the attachment to Dropbox. Then it sends a Slack message with a link.⁴

Nobody opens the email, saves the file or writes the message. But somebody still had to choose each of those jobs and the order they run in.

What you need before you build one

• A task you can say in one sentence, with a clear start. "When a new lead arrives" works. "Handle my sales" doesn't.
• The apps involved, and an account in each.
• A list of the jobs in order.
• An idea of what should happen when something's missing, such as an email with no attachment.

n8n states a rule worth knowing early. All production workflows need at least one trigger to determine when the workflow should run.²

Where the counting starts

The makers count usage in different ways. Make counts an operation as a single module run to process data or check for new data.³ Zapier counts tasks, and its Dropbox and Slack example uses two tasks per run.⁴

In Make, the count rises with the items processed, so a run with many items can use more than a run with one.³ A later page lines the three makers up in one table: Zapier, Make and n8n.

Where an AI step fits

Some workflows add a step where a language model reads text and writes something back. It's one more step in the chain. It still gets its input from the step before and passes its answer on. What an AI step does inside a workflow explains what that step does, and where a person should check it.

One warning before you add any AI step: passwords, API keys, customer records and card numbers stay out of it, unless your business has a written rule that allows otherwise.