The best way to automate a repetitive business task depends on what the task is made of, not how often it repeats. Rule-shaped work belongs in a trigger. Work that needs judgment breaks rules and needs something that can read a situation. Some of it should just be deleted.
We build software for owners who run their businesses from a truck and a phone. The pattern we see is not that they automate too little. They hand the wrong task to the wrong kind of tool, then decide automation does not work at their size.
| The mechanism | Use it when the task is | Setup | What it costs to keep alive |
|---|---|---|---|
| Kill it | Only there out of habit: a form, a step, a report nobody reads | Minutes | Nothing |
| Template it | The same words every time, written from scratch anyway | One sitting | Effectively nothing |
| Trigger it | Rule-shaped: same input, same output, no exceptions | An afternoon | A monthly fee, plus the maintenance nobody budgets for |
| Hand it to something that decides | Judgment-shaped: needs reading the situation before answering | Varies | Supervision rather than maintenance |
Why the last thing you set up quietly stopped working
Work you do by hand fails loudly. You forget the supplier form, the supplier calls, and you know within a day.
Work you hand to a rule fails quietly. The app on the other end changes what it sends, a login expires, or something arrives in a shape the rule was never told about, and the rule does its job with nothing. Nothing announces that it stopped. You find out weeks later, from a customer.
That gap between the failure and the discovery is where the real cost of automation sits. Which is why the useful question is not which task to automate, but what you are handing it to.
How often it repeats tells you which task. What it’s made of tells you what to use.
Nearly every guide gives the same exercise: list what you did more than twice last week, multiply minutes by frequency, automate the top three. That finds the target and says nothing about the weapon. Two tasks that repeat as often can need opposite mechanisms.
The businesses that have adopted this stuff are not automating everything either. Among US firms using AI, 57% keep it inside three or fewer parts of the business, and among those using it on specific tasks, 65% keep it to three or fewer (Source: US Census Bureau, CES-26-25, 2026). That is restraint in national data, from firms with more staff and budget than anyone reading this.
If you want the list of what can be automated at all, a full inventory of what’s actually automatable has it, and which task to start with handles the ordering.
The two mistakes that cost the most
The first is paying every month for variability that is not there. A deposit reminder says the same thing to every customer at the same point in the job. Handing that to something clever is renting judgment for a task with one right answer. A template does it for nothing.
The second costs more: wiring a rule for work full of exceptions. The reminder goes out, one customer asks for Thursday instead, and the rule has no answer. So you add a branch, then another, and every branch is maintained by you. Eventually the automation is a second job you did not price.
Both are the same error pointed in opposite directions. The mechanism did not match the shape of the task.
How to tell which one a task needs
Four questions, answerable standing next to a van.
- If you stopped doing this entirely, who would notice? If the answer is nobody, you are done. Do not automate it, delete it. (Cutting the admin before you automate it is a discipline of its own.)
- Is the output the same words every time, or does it depend on what they said? Same words every time means a template. The composing is the work, and the composing is what you remove. Text Replacement on an iPhone or a saved reply in Gmail does this tonight, for nothing.
- Can you write the rule down in one sentence with no “unless” in it? If yes, trigger it. Keep it to one trigger and one action, which is the shape that fits inside a free plan. Every “unless” is a branch you will be maintaining next spring.
- Does doing this well require reading the situation first? Then it is judgment-shaped, and that is the hard case.
Question three does the work. Say the rule out loud.
If an “unless” shows up, you have not found a rule. You have found a decision wearing a rule’s clothes.
| Rule-shaped work | Judgment-shaped work |
|---|---|
| Same input, same output | The right answer depends on what the customer said |
| Writes down in one sentence | Every attempt to write it down grows an “unless” |
| Fails silently when an input changes shape | Fails visibly: a wrong answer, not a missing one |
| A trigger holds it for roughly $9 to $20 a month | Needs a person in the loop |
What this actually costs in the month it’s working hard
Price the busy month, because the quiet month will lie to you.
Take one genuinely rule-shaped job: a booking confirmation when the job is set, a reminder the day before. Two messages per job, no exceptions. At four jobs a day that is eight triggered actions daily, and Zapier’s free plan stops at 100 tasks a month (Source: Zapier pricing, 2026), gone before the second week. Make’s free plan allows more volume but caps you at two active scenarios and will not run them more than every fifteen minutes (Source: Make pricing, 2026), which is useless when the point is a same-day confirmation.
So the real number for a task worth automating is the paid entry tier: about $9 a month at Make, about $19.99 a month billed annually at Zapier.
An automation worth building costs ten or twenty dollars a month and hands back the evening you would have spent doing it by hand. An automation not worth building costs the same ten or twenty dollars and then takes a different evening every month to keep alive. The price is identical. The shape of the task is the whole difference, which is why the mechanism matters more than the tool. (How few tools you actually need makes the same case about the stack.)
What to do with the judgment-shaped work right now
The honest position: for back-office work, the tool that reads a situation and decides well is arriving rather than finished, and telling you to wait for it is a poor answer to a problem you have this week.
So hold that work in a template plus a fixed time to run it. The template removes the composing. You keep the deciding, which is the part that needed you. It costs nothing to maintain and does not break silently, because you are the one running it.
What it does not do is scale, and you will feel that. The national pattern is the same shape: 66% of firms using AI use it to assist rather than replace, and the leading uses are writing, document analysis and search (Source: US Census Bureau, CES-26-25, 2026). Drafting work, with a person still deciding.
What Beside does and doesn’t do here
Beside today is a phone product. It answers calls and texts, transcribes them, and sends summaries with suggested actions. It does not do invoicing, bookkeeping, data entry, or general task automation, which means it does not do most of what this article is about.
If you came here to get a repetitive task off your plate this week
The next step does not need us. Take your own list this week and run the four questions down it: kill what nobody would miss, template what comes out the same words every time, trigger the ones you can state in a sentence with no “unless” in it. Whatever is left is judgment-shaped, and that is the pile worth putting on a waitlist instead of building a rule you will still be repairing next spring.
Sources: US Census Bureau, CES-26-25, US Census Bureau AI-use data, Zapier pricing, Make pricing. Beside features and pricing verified on beside.com/pricing on August 6, 2026. Photo: Pexels.