Automating Repetitive Business Tasks: Exceptions First

Most guides tell you which task to automate. Few tell you what the automation should do when the task goes sideways. Here is the exception rule I write first, with a lead follow-up example.

It is 4:40 on a Tuesday afternoon. A new lead lands in your inbox from the website form. You copy the name into a spreadsheet. You paste the same name into your CRM. You type a quick "thanks, I'll call you tomorrow" email. Then you remember the two forms that came in this morning, and neither of them ever got that email.

None of this is hard. That is the problem. It is dull, it never stops, and it quietly eats the hours you meant to spend on actual work.

So you do what every guide tells you. You pick a task, you pick a tool, and you build something. Most of those guides, including big vendor explainers like IBM's overview of task automation, walk you through choosing what to automate. That part is fine. I want to talk about the part they skip: what the automation does on the day the task stops looking like the demo.

The Task Is Rarely the Problem. The Exceptions Are.

Every repetitive task has a happy path and a pile of weird cases. The happy path is what you describe when someone asks what you do. The weird cases are why you still do it by hand.

Why the demo always works

A tutorial uses clean data. One form, one name, one phone number, one tidy result. Your real inbox has a lead who typed a letter O instead of a zero, a customer who submitted twice, and a spam bot that found your form at 3 a.m.

If the automation has no plan for those, it does one of two things. It fails quietly and nothing happens. Or it charges ahead and sends a cheerful text to the wrong person. Both are worse than doing it by hand, because when you did it yourself, at least you noticed.

Write the Exception Rule First

Before I open any tool, I write one sentence for the happy path. Then I write one line per exception. Each exception gets exactly one of three fates:

Handle it. The automation knows what to do and does it.
Park it. The item goes into a list that a human checks on a schedule.
Alert a human. Something pings you right away, because waiting costs money.

That is the whole framework. It takes ten minutes with a pen, and I almost never see anyone write it down.

A worked example: lead follow-up

Here is the happy path. When a form arrives, send a text within minutes, add the lead to the CRM, and notify the owner.

Now the exceptions:

1. No phone number, or a number that fails a basic format check. Park it, and send an email instead if there is one. 2. Same person submitted twice. Handle it. Merge into one record and send one message. 3. Arrives at 11 p.m. Handle it. Queue the text for the morning and say nothing about the delay. 4. The email domain matches an existing customer. Alert a human. This is not a new lead, and a canned reply would be awkward. 5. It looks like spam. Park it. Never delete anything you cannot get back.

Five lines, and only one of them wakes you up. If every exception pinged me, I would switch the whole thing off within a week. The real skill is deciding which problems deserve an interruption.

Start With the Boring One

Most small businesses that ask me about automation have the same three problems. Leads do not get followed up fast enough. The same data gets typed into two systems. Calls go unanswered. Not one of those is a chatbot.

The first automation that pays for itself is usually the boring one, and fast lead follow-up is the best candidate. People have studied why speed matters here for years. A well-known Harvard Business Review analysis of online sales leads found that responding quickly made a big difference to whether a business reached the lead at all. It is an older study, but the logic holds up. People contact several providers and go with whoever answers first.

The unanswered call is the same problem, only louder. I built CallCrewHQ because a tradesperson cannot pick up the phone from under a sink. The exception rule mattered there more than anywhere. An AI front desk that answers every call has to know when to take a message, when to book, and when to put a caller straight through to a person. I wrote those handoff rules before I touched the voice side.

I keep a running list of the other boring wins, and the ones that flopped, on the front page of this blog.

Decide Where It Runs, and Who Owns It

This question comes next, and it rarely gets an answer. Once the automation exists, where does it live and who looks after it?

I work with n8n, Supabase, Next.js, Docker on a VPS, and the Anthropic API. I run all of them in production, so I can give you the trade-off plainly. Self-hosting n8n on a VPS is cheap and reliable. But somebody has to own the updates, backups and secrets. A managed tool takes that job off your hands and charges a monthly fee for the privilege.

Neither answer is wrong. Not choosing is wrong. If you self-host, put three dates in your calendar: one for updates, one for testing a restore from backup, and one for rotating keys. The n8n hosting documentation covers the mechanics. If nobody on your team will act on a reminder like that, pay for the managed version and call it what it is: a maintenance contract.

I go into this in more depth on my blog about building automation in public, where I write about what breaks.

What Happens to the Manual Way Afterward

Nobody tells you this part. Once the automation goes live, do you retire the old process?

Not yet. I run both for two weeks. The automation does the job, and I spot check five items a day by hand alongside it. If my check and the automation disagree, the rule is wrong, and I fix the rule rather than the item.

Then I add one standing habit. Once a week I spend ten minutes reading the parked list. If it holds a handful of odd cases, the system is healthy. If it keeps growing, one of my exceptions is really a common case in disguise, and it deserves its own handling.

That habit builds more trust than any dashboard. You stop wondering whether the thing is working, because you have a place where its doubts pile up.

Your Next Ten Minutes

Pick one task you did three times this week. Write the happy path in one sentence. List every way it has gone wrong before. Give each one a fate: handle, park, or alert. If you cannot write the exception list, you are not ready to automate that task yet, and that is worth knowing.

Small businesses make up the vast majority of U.S. firms, according to the SBA Office of Advocacy, and most of them are short on exactly one thing: hours. Handing the dull, constant work to a machine is the right instinct. Just give it rules for the bad days.

If you want a hand writing that exception list or building the thing, get in touch through my site. Tell me the task, and I will tell you honestly whether it is worth automating.

---

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

Automating Repetitive Business Tasks: Write the Exception Rule Before You Build