When a team decides to automate something, the list of candidates gets long quickly. Replying to enquiries, chasing invoices, writing reports, updating the CRM, sorting the inbox. Everything looks like a good idea, so nothing gets started.
The first one matters more than it seems. If it goes well, the team trusts the next one. If it goes badly, everyone gets cautious. So it is worth choosing carefully, and the choice is usually less exciting than people expect.
Start with the repeated task, not the impressive one
The best first workflow is the one people do again and again and quietly dislike. It might take ten minutes each time, but it happens fifty times a week. That adds up, and it is easy to see whether the automation worked because the old way is still fresh in everyone's mind.
A rare, complicated task might look more valuable on paper. It is also harder to test, because you will not see it often enough to know whether the result is right.
Three questions that narrow the list
How often does it happen? Daily is better than monthly, because you learn faster.
What does it cost when it goes wrong? Drafting a reply that a person reads before it goes out is cheap to get wrong. Sending money or changing a customer record is not. Begin where mistakes are small and easy to catch.
Is the answer already written down? If the right outcome lives only in one person's head, an agent will struggle. If there is a document, a price list or a set of rules, it has something solid to work from.
Pick something with a clear finish line
It helps if you can say exactly when the task is done. A booking is confirmed. An invoice reminder is sent. A report lands in the right inbox. Vague goals such as "improve customer experience" make it hard to tell whether the automation is doing its job.
Keep a person in the loop at first
For the first few weeks, let someone check what the automation does before it goes out. This is not a lack of faith. It is how you find the cases nobody thought of. Once the checks stop turning up problems, you can step back and let it run on its own for that part of the work.
After the first one works
Write down what you learned. Which inputs were messy, which rules turned out to be unwritten, where people disagreed about the right answer. That list becomes the starting point for the second workflow, and the second one is almost always quicker than the first.