Automate Repetitive Business Tasks: Count First

A guide for owners who suspect they lose hours to the same chores every week. By the end you can decide which task to automate first, which tool class fits, and whether to do it yourself.

This is for the small business owner or operator who keeps doing the same chores by hand and wonders whether a tool could take them over. By the end you'll be able to decide three things: which task to automate first, what kind of tool fits it, and whether to build it yourself or pay someone else to.

I'll skip the usual list of ten tasks every business should automate. Those lists are fine, but they hand you my chores, not yours. The better question is which of *your* chores deserve a machine. You can answer that with a pencil and one week.

Count the Work Before You Pick a Tool

Most people choose a tool first and then go hunting for something to feed it. That's backwards. It's also how you end up with three subscriptions and the same Friday afternoon of copy-pasting.

Try a tally week instead. Grab a sheet of paper, or open a note on your phone. For five working days, write one line every time you do something you've done before. Keep it dull. Note the task, roughly how long it took, and what triggered it. A lead came in. An invoice got paid. A customer asked for a booking confirmation.

By Friday you'll have a column of repeats, and three things usually jump out.

Some tasks repeat far more often than you thought. You never noticed because each one takes two minutes.
Some tasks are huge but rare. The monthly report feels awful, but it happens once.
Some tasks have the same trigger every time. These are your candidates, because a machine can only act on something it can detect.

That last point matters most. If you can say "every time X happens, I do Y," you have an automation. If you can't name X, you have a judgment call, and it should stay with a human for now.

The Three Chores That Show Up Again and Again

I run my own products. CallCrewHQ answers phones for trades. Grease Trap Quotes is a quote aggregator for commercial kitchens. Course Advisor handles voice enrollment for education leads. I build with n8n, Supabase, Next.js, Docker on a VPS and the Anthropic API. That mix has taught me that most small businesses asking about automation have the same three problems.

Leads that don't get followed up fast enough.
Data typed twice into two different systems.
Calls that go unanswered.

If your tally sheet shows one of these, start there. And notice what's missing from that list. Nobody's first problem is a lack of chatbot. The first automation that pays for itself is usually the boring one, like replying to a new lead within minutes. Someone who asked about your service an hour ago is already talking to your competitor.

The case for speed is old and well documented. A Harvard Business Review study, The Short Life of Online Sales Leads, looked at how quickly companies responded to web leads. It found that waiting made those leads much harder to reach and qualify. The research isn't new, but the human side of it hasn't changed one bit. People want an answer while the question is still fresh in their heads.

Four Questions Before You Automate Anything

You now have a column of repeats. Run each one through these four questions. I treat them as a gate, not a score. A task has to pass all four.

1. Can you name the trigger in one sentence?

"When a form is submitted" passes. "When a customer seems unhappy" doesn't. If the trigger is a feeling, wait.

2. Do the steps change from one run to the next?

Run the task in your head for three different customers. If the steps are identical, good. If you keep saying "well, it depends," count the exceptions. When more than about one run in five needs a human decision, automate only the front half and hand off at the fork.

3. What happens if it quietly goes wrong?

Most guides skip this one. Every automation fails eventually. A bad reminder email is annoying. A bad invoice is expensive. Decide now who sees the failure and how. If nobody would notice for a week, you need an alert before you need the automation. I wrote more on this in what happens when a tool breaks quietly. The short version: silence is the dangerous outcome.

4. Would you still do this task by hand if the tool vanished tomorrow?

If the answer is no because you've forgotten how the process works, you're building a dependency, not a convenience. Write the steps down in plain words before you build. That page doubles as your spec and your insurance.

A Worked Example: The Missed-Lead Decision

Say you run a small HVAC company with two office staff. Your tally week shows four repeats.

Typing job details from a web form into your scheduling software.
Texting customers an appointment reminder the day before.
Chasing unpaid invoices.
Calling back web leads, whenever someone gets a gap.

The tempting pick is invoice chasing, because it's the most annoying. Run it through the four questions and it stalls. The trigger is fuzzy, since some customers pay late for a reason, and the wrong reminder to the wrong person damages a relationship. Question two catches it.

The web-lead callback passes every gate. The trigger is a form submission. The steps never change: an acknowledgment text with a link to book. If it fails, you'll see it, because the lead stays unbooked. And the cost of the current process is clearest here. Every slow callback is a job that might go to someone quicker.

So you build one thing. A form submission fires off a text and an email in under a minute, and a row lands in your scheduling system so nobody has to retype it. That handles the lead response and the double entry in one go. Invoice chasing can wait.

That's the rule in action. Pick the task with a clean trigger and an obvious cost of delay, not the one that annoys you most.

Your Options, Compared

Once you know the task, you choose how to build it. Here are the realistic routes and what drives the cost of each. I've left out prices on purpose, because they change and they depend on your volume.

Do nothing but write a checklist. Costs only your time. Right for tasks that are rare or full of exceptions. It won't stop the typing, but it stops the forgetting.
Use the features already in your software. Most CRMs, schedulers and accounting packages have built-in reminders and rules. The cost driver is your plan tier. Check this before you buy anything else, because plenty of owners pay for a tool and never switch the feature on.
A hosted no-code tool. Quick to start, with billing that usually scales by the number of runs. The cost driver is volume, so a task that fires thousands of times can get pricey. I compare the main options in how I'd choose between Zapier, Make and n8n.
Self-hosted n8n on a VPS. Cheap and reliable to run, and it's how I run mine. The catch is that somebody has to own updates, backups and secrets. If that somebody is you, budget an hour a month and write down where the credentials live. The n8n hosting docs show what setup involves.
Custom code. Full control and the highest build cost. Only worth it when the process *is* the product, or when no existing tool can see your trigger.
A contractor or advisor. You pay for their time instead of spending yours. The cost driver is how clearly you can describe the task, which is one more reason to finish the tally week first.

The cheapest route that passes all four questions is the right one. Complexity is not a prize.

What It Costs to Own, Not Just to Build

Building is the cost you can see. Owning is the one people forget. Every automation needs three things after launch.

An owner

One named person gets the alert when it breaks. If the answer is "whoever notices," the answer is nobody.

A backup

If your workflows live in a tool, export them. If your data sits in a database, check that backups actually run and that you can restore one. Supabase explains how this works in its backup guide, and the same discipline applies to anything you store.

A review date

Businesses change. Put a date in the calendar three months out and ask whether the automation still matches how you work. Half the failures I've run into weren't bugs. The process changed and the automation didn't.

For the longer version, I broke down what a tool really costs to own in a separate post.

Where AI Fits, and Where It Doesn't

A language model is good at reading messy text and working out what it is. Is this incoming email a quote request, a complaint or spam? That's a real job for a model, and the Anthropic API documentation is a fair place to see what's possible.

It's a poor choice for anything with a fixed rule. If every invoice over a set amount needs a second approval, write that as a plain condition. Rules are cheaper, faster and never have an off day. Use a model when the input is unpredictable but the rule for the output is easy to write.

What to Do on Monday

Here's a plan small enough to start before lunch.

Start the tally. One line per repeat, all week.
On Friday, circle every task with a clean trigger.
Run the circled ones through the four questions.
Pick one. Write its steps in plain words.
Choose the cheapest route from the options above.
Name the person who gets the alert when it breaks.

That sequence beats any amount of tool shopping. One working automation with an owner is worth more than five that nobody remembers.

If you want to see how I think about this in practice, the rest of my writing lives on my blog. And if you've got a repeat task you can't make up your mind about, tell me what triggers it and what happens when it goes wrong. I'm happy to talk it through and give you an honest view on whether it's worth automating at all.

---

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

Automate Repetitive Business Tasks: Count First, Then Pick What to Hand Off