Automation Tasks: 7 Steps to Pick the Right Ones

Most people start with the task that annoys them most. I start with the one that's boring, frequent, and cheap to get wrong. Here are seven steps in the order I'd do them.

The most common mistake with automation is picking the task that annoys you most. The annoying task is usually rare, messy, and full of judgment calls. The good candidate is the dull one you do every day without thinking.

Here are seven steps, in the order I'd do them.

1. Write each task as one plain sentence

Open a notes app and describe the task like this: "When X happens, I do Y, and then Z gets updated." If you can't finish that sentence, the task isn't ready. You've found a judgment call, and judgment calls come later.

This sounds too simple to matter. It matters because the sentence names your trigger, and every automation starts from a trigger. If you want more on why, I wrote about it in the trigger is the whole automation.

Good: "When a web form comes in, I copy the name and phone number into the CRM."
Not ready: "When a customer seems unhappy, I handle it."

2. Count how often it happens and how long it takes

Guessing is how people end up automating a four-minute-a-week job. Count for one real week. Tally marks on paper work fine.

You need two numbers. How many times did it happen, and roughly how many minutes did each one take? Multiply them. A task that takes 30 seconds but happens 40 times a day will beat a task that takes an hour once a month. Nobody believes this until they see the tally.

3. Score the cost of getting it wrong

This is the step most checklists skip. Ask what happens if the automation does the wrong thing and nobody notices for a week.

Here's a worked example with two candidates. One sends a "thanks, we got your inquiry" text to every new lead. The other sends invoice reminders to customers who owe you money. Both happen often. If the text goes out twice, the lead rolls their eyes. If the reminder goes to someone who already paid, you've annoyed a customer and maybe lost them.

So you build the text first. You save the same time, and the mistakes cost far less. Rule of thumb: start where a wrong run costs you an eye roll, not a refund.

Low cost of error: internal notifications, confirmations, copying data into a spreadsheet.
High cost of error: anything that sends money, deletes records, or speaks for you to an angry customer.

4. Check it against the three usual suspects

Most small businesses that ask me about automation have the same three problems. Leads don't get followed up fast enough. Data gets typed twice into two systems. Calls go unanswered.

Hold your list up against those three. If a task fits one of them, move it to the top. The first automation that pays for itself is usually a boring one, like following up with a lead within minutes. It's rarely a chatbot, no matter how much the chatbot demos look like the future. I keep notes on what I've built and what broke along the way, if you want the unglamorous version.

5. Pick the tool by who will maintain it

Skip the feature comparison for now. Ask who will fix it at 7 a.m. when it stops. A managed tool costs more each month, but somebody else owns updates and uptime. Self-hosting costs less and runs well, but you own the updates, the backups, and the secrets.

I work in n8n, Supabase, Next.js, Docker on a VPS, and the Anthropic API, so I've run the self-hosted route in production. It's cheap and reliable. It also means someone has to care about it. The n8n hosting docs lay out what self-hosting involves, and they're worth reading before you decide. If nobody in your business wants that job, pick the managed option and don't feel bad about it.

6. Build the smallest version and run it beside your hand work

Build only the trigger and one action. Leave out the clever branches. Then run it alongside your manual process for a few days and compare the results.

This catches the awkward stuff early, like a form field that sometimes arrives empty or a phone number in a strange format. Fix those one at a time. Add the second action only after the first has behaved for a week. And don't turn on anything that touches customers until you've read the output yourself.

The Small Business Administration has general guidance on running a small business. It's a reasonable place to check that what you're automating fits how you actually operate day to day.

7. Decide how you'll find out when it stops

This is the other step nearly everyone skips. Automations rarely fail with a loud crash. They go quiet. A login expires, a field gets renamed, and the thing stops doing its job while everything looks calm.

Pick one signal that tells you it's alive. A daily count works. So does a message to your phone that says "12 leads handled today." If that message doesn't arrive, you know something broke. The missing message is the alarm.

Also write down three things: where the credentials live, where the backup lives, and who to call. CISA's Secure Our World guidance covers the basics of passwords and updates, and it applies to your automation accounts as much as to your email.

Where to go from here

Pick one task from your list. Run it through steps one to four tonight. By tomorrow you'll know which one to build first.

I build and run my own products on this stack, including CallCrewHQ, Grease Trap Quotes, and Course Advisor. If you'd like a hand choosing your first automation task, get in touch through my home page and tell me what you're doing by hand right now. No pitch. I'll just tell you what I'd build first.

---

*Written by Mario Lucas. Mario runs Utomat, an automation studio that builds AI voice agents and workflow automation for small businesses.*

Automation Tasks: 7 Steps to Pick and Start the Right Ones