Most automation advice starts with a tool and works backwards to a problem. That is how businesses end up paying monthly for four things that overlap and solve nothing. Start with the jobs instead.
The test that actually works
A job is worth automating when it is repetitive, rule-based and low-stakes if it goes slightly wrong. All three. Miss one and you either waste the effort or create a problem worse than the one you started with.
Repetitive means weekly or more. Automating something you do twice a year costs more in setup and maintenance than it will ever return. Rule-based means you could write the steps down for someone else — if the decision needs judgement about a particular customer, it is not rules, it is you. And low-stakes matters because every automation eventually misfires; the question is whether that costs you five minutes or a client.
Four worth doing
1. The monthly report nobody writes
Pulling analytics into something readable is repetitive, rule-based, and if the wording is slightly off nobody is harmed. It is close to the perfect candidate, which is why it is the one we automated first in Insights — GA4, Search Console and Google Ads pulled together and drafted, with you approving before it goes anywhere.
2. Knowing when something breaks
Checking whether your website is up is the definition of a job a computer should do. A human checks when they remember; monitoring checks constantly and only speaks up when something is wrong. How monitoring works →
3. Backups and renewals
Backups, SSL renewals and domain renewals are pure rules with a nasty failure mode when forgotten. Automate them, then — and this is the part people skip — test that a restore actually works. An untested backup is a rumour.
4. The first draft of anything written
Product descriptions, service pages, the post you have been meaning to write. Not the final version — the blank page. Getting from nothing to something rough is the slowest part of writing and the part a machine is genuinely good at.
Three that are not worth it
Replying to customers
Fails the low-stakes test badly. An automated reply that misreads a complaint turns a small problem into a large one, and the time you save is measured in minutes against a cost measured in customers. Route and prioritise automatically by all means. Let a person write the reply.
Anything you do rarely
Quarterly and annual jobs almost never repay the setup, and worse, an automation you only see four times a year is one you will have forgotten how to fix when it breaks.
Decisions you would need to explain
If you would struggle to justify a decision to a customer, an accountant or a regulator, do not hand it to a system that cannot explain itself either. Pricing, hiring, and anything touching someone's money or personal information belong to a person.
Start with one
The common failure is automating six things in a fortnight, understanding none of them, and quietly abandoning the lot. Pick the job that annoys you most this month, automate that, and live with it for a few weeks before adding another.
If you want the wider picture first — what AI is good at, what it is not, and where the two overlap with automation — that is covered in AI for small business, without the hype.
Want the reporting half of this handled for you?
See Insights →