Automation feels productive, which is exactly why it is easy to do badly. Before you build or buy anything, answer these five questions on paper.
1. What exactly triggers the process?
Every automation starts with an event: an email arrives, a form is submitted, a payment lands. If you cannot name the trigger, you cannot automate the process.
2. What are the steps, in order?
Write them down as a numbered list, including the "obvious" ones nobody mentions. Hidden steps are where automations break.
3. Where is a human decision really needed?
Some steps need judgement: approving a refund, replying to an upset client. Mark them. A good automation does the preparation and hands the decision to a person.
4. What happens when it goes wrong?
Tools fail, emails bounce, models misread a sentence. Decide in advance who gets notified and how the work continues by hand. If the answer is "nobody knows", the process is not ready.
5. How will you know it is working?
Choose one number before you start: minutes saved per week, replies sent within an hour, invoices sent on time. Check it after two weeks.
A quick example
A small shop wanted to "automate customer service". Using these questions we narrowed it to one thing: order-status questions, which were 40 percent of all messages. The trigger was an incoming message, the steps were looking up the order and replying, the human decision was refunds, and the success number was first reply time. Two weeks later the owner had her mornings back.
Automate one clear thing, measure it, then do the next.
