Which admin tasks can a small business actually automate?

What makes an admin task worth automating?

Three tests, and a task has to pass all three.

It repeats. The same job, done the same way, again and again. Sending the weekly timesheet reminder repeats. Sorting out a supplier dispute does not.

It follows a rule you could write down. If a new starter could do it from a written list of steps, without once having to ask what you meant, a machine can follow the same list. If the honest explanation includes "you get a feel for it", it fails.

Then frequency. A job done once a year is usually cheaper to keep doing by hand than to automate, test and look after. A job done every morning is not.

Test	Ask yourself	Passes	Fails
Repetitive	Is it the same job every time?	Chasing the same timesheets each week	Handling a complaint
Rule based	Could a new starter follow written steps?	Filing invoices by supplier name	Deciding which customer comes first
Frequent	Does it happen weekly or more often?	Copying form responses into a tracker	The annual insurance renewal

Most small businesses have a handful of jobs that pass all three, and they tend to be the ones people describe with a sigh. Copying enquiries from the website form into a spreadsheet. Renaming and filing attachments. Sending the same reminder every Thursday. Building the same summary from the same three tabs every Monday morning.

Nobody started a business to do any of that.

What can the tools you already have actually do?

More than most people use. If the business runs on Google Workspace, two features cover a lot of everyday admin.

The first is Gmail filters. Google's help page on filters explains that you can use them to send email to a label, or to archive, delete, star or automatically forward it. No code involved. If every supplier invoice arrives from a known address, a filter can label it and move it out of the inbox before anyone has to look. One catch: if the filter forwards mail, Google says only new messages are affected, so it will not forward the backlog.

The second is Apps Script, which lets you write short scripts that work with Sheets, Docs, Drive, Gmail and Calendar. The part that matters for admin is triggers. Google's guide to installable triggers describes clock triggers, which run a script at a set time or on a repeating interval, anywhere from every minute to once a month. It also describes event triggers, which run when someone submits a form or edits a spreadsheet.

That is the difference between a task someone has to remember and a task that simply happens. The Thursday reminder sends itself on Thursday. Form responses land in the tracker as soon as they are submitted. On Monday the summary is built before anyone logs on.

We build this kind of thing in our Google Apps Script service. The wider version, across tools that are not Google, is process automation.

Where do people get automation wrong?

The usual mistake is automating the mess. A process with five workarounds becomes an automated process with five workarounds, running faster, and soon nobody remembers why the workarounds exist. Write the steps down first. If writing them down is hard, that is the real job, and it comes before any tooling.

The second is building everything under one person's login. Google's guide is plain about it: installable triggers always run under the account of the person who created them. That works fine until that person changes role or leaves. Run the automation from an account the business controls, and write down where it lives and what it does.

The third is treating silence as success. When a triggered script fails, the same guide says no error message appears on screen. Apps Script sends a summary of failures by email instead. If nobody reads that email, a broken automation can sit there for weeks while everyone assumes the job is being done.

The fourth is ignoring the limits. Free tools have ceilings. Google's quotas page says that if a script exceeds a quota or limitation, it throws an exception and execution stops. Total trigger runtime is capped at 90 minutes a day on a consumer account such as gmail.com and 6 hours a day on a Google Workspace account, and quotas reset 24 hours after the first request. Google also says those numbers can change without notice, so check the page rather than relying on a blog post. A business sending a few reminders a day is very unlikely to get close. A script that reworks every row of a large sheet every minute might.

How do you find the admin worth automating?

Count it for one week. Keep a plain list: each admin task, how often it came up, roughly how long it took, and whether you could write the steps down. Do this as you go, not from memory at the end of the week. Memory is generous to the jobs we like and harsh on the ones we hate.

At the end of the week, sort the list. Anything frequent and rule based goes to the top. Anything that needed judgement comes off, however long it took.

Task	How often	Written steps?	Verdict
Copy web enquiries into the sheet	Daily	Yes	Automate
Chase missing timesheets	Weekly	Yes	Automate
Reply to a complaint	When it happens	No	Keep it human

Add your own column for the time each task took. Your figures are the only ones that count here.

What comes out of this is rarely one big project. It is usually a few small jobs, each of which can be dealt with on its own, and the first one tells you whether the rest are worth doing.

When is automation not the answer?

When the task needs judgement. Deciding whether to give a customer more credit, how to word a difficult email, or which complaint to deal with first. A machine can gather the information for those decisions. It should not be the one making them.

When the decision is about a person and matters to them. The ICO's guidance on automated decision making describes it as using personal information to make a significant decision about someone using solely automated processing, and says the UK GDPR sets out specific requirements that apply to it. If an automation would decide something significant about a customer, an applicant or a member of staff with no person checking it, read that guidance before you build anything, and take proper data protection advice. Keeping a person in the loop is usually the simpler answer.

Rare jobs are the third case. The setup, the testing and the upkeep cost more than the hand work they replace.

When nobody can write the steps down. Then the job is to agree the process first. A script will not fix a process nobody has written down, and neither will AI.

And sometimes the right answer is to stop doing the task altogether. A report nobody reads does not need automating. It needs deleting.

Where should you start?

Pick one task from the week's list. The most boring one that passes all three tests is usually the right first choice, because it is easy to check that it works and obvious when it stops.

If you have the list and are not sure what is worth building, tell us what's stuck and we'll map the quickest fix. We will tell you straight which jobs a filter or a short script can handle, which need something bigger, and which should stay with a person. Time to talk yet?