The best automation targets are boring, frequent and rule-based. The worst are rare, high-stakes and judgement-heavy.
Quick answer
A task is worth automating when it happens often, follows rules you can write down, is cheap to check, touches data you are allowed to use, and is unlikely to change next month. If it is rare, has serious consequences when it goes wrong, or depends on judgement that you cannot describe, keep a person in charge of it. Everything in between can often be automated with a person checking the output.
Five questions to ask about any task
- How often does it happen? A task done daily or weekly repays the effort. A task done twice a year usually does not.
- Can you write down the rules? If you can explain how a good result is chosen in a few clear steps, it can be automated. If the answer is "it depends, I just know", it cannot yet.
- What happens if it goes wrong? A mislabelled email is a nuisance. A wrong invoice or a message sent to the wrong customer is a problem. The higher the cost of a mistake, the more a person needs to check.
- What data does it touch? Personal, financial or confidential data needs a clear reason and a safe place to be processed before it goes near any AI tool.
- Will the task stay the same? Automating a process that is about to change just builds something to throw away.
Good candidates
- Sorting and routing incoming messages, such as enquiries, supplier emails or form submissions
- Copying data between tools, where someone currently retypes or pastes the same details
- Building recurring reports from the same sources every week or month
- Checking items against a list, such as whether a document includes everything required
- Drafting a first version of a routine response for a person to edit and send
Poor candidates
- One-off decisions with no pattern to learn from
- Sensitive conversations with customers or staff
- Anything legal, financial or regulated that nobody checks before it goes out
- Processes that are broken. Automating a messy process just makes the mess faster. Fix the process first.
Score it before you build it
For each task you are considering, write down three things: how many times it happens a month, how long it takes each time, and what a mistake would cost. Multiply the first two to get the time spent. A task that costs hours a month and has a low cost of error goes to the top of the list. A task that costs minutes and has a high cost of error goes to the bottom.
This is the same logic used to prioritise any list of work, as described in how to turn an SEO audit into an action plan: rank by impact and effort, then do the best few first.
Common mistakes
- Starting with the tool and looking for something to use it on
- Automating the most annoying task instead of the most time-consuming one
- Removing the human check because the first few results looked fine
- Not measuring afterwards, so nobody knows whether time was actually saved
- Putting personal data into AI tools without checking where it goes and how it is stored
When to get support
If you have a list of tasks and are not sure which are worth the effort, an AI Automation Review ranks them by time saved, effort and risk, and ends with a plan for what to build first.
FAQ
How do I know if a task is too risky to automate?
Ask what the worst realistic mistake would cost, and whether anyone would catch it before it reached a customer. If the answer is "a lot" and "no", keep it manual or add a person to check the output.
Do I need technical skills to automate a task?
Not always. Many workflows can be built with tools you already use, such as Google Workspace, spreadsheets and forms. More complex ones need someone to build and test them.
Should I use AI for everything I automate?
No. Plain rules are often cheaper, more predictable and easier to check. AI is useful where the input is messy, such as free-text email, and rules alone cannot cope.