On this page
The decision to automate repetitive tasks starts with the wrong question in most businesses: “which tool should we use?” The right question is: “is this task actually worth automating?” Not every repetitive task suits automation, and forcing one that doesn’t costs more than doing it by hand.
Below is the five-question filter I put a task through before it goes on an automation list. A task that gets a yes to all five is a strong candidate. If it gets a no to one or more, it either needs fixing first or should come off the list. You can try the same filter on your own task with the automation calculator.
Why automation decisions start in the wrong place
When a team raises the idea of automating repetitive tasks, the work that lands on the table is usually the most visible. And the most visible work is usually the most complained-about, not the most time-consuming.
Those are different things. A task done once a week that everyone hates gets talked about far more than one done quietly three times a day. If automation investment follows the complaints, the time saved falls well short of expectations and the project gets labelled “didn’t work”.
The filter exists to break this emotional prioritisation. The questions are simple, but asked in order they eliminate half the list.
The five-question filter
The filter turns the question of which tasks can be automated into a concrete checklist. Ask the questions in order. When a task stops at one of them, note why before taking it off the list, because that note usually shows where the real problem is.
1. Is the task repeated regularly, always in the same way?
The return on automation comes from repetition. A task done once a month usually doesn’t cover the cost of building and maintaining it. One done several times a week does.
“In the same way” matters as much as the repetition. If the task runs slightly differently each time, there’s no standard process. Automating a non-standard process does nothing except speed up the uncertainty.
2. Can the steps be written down?
The practical test: could the person doing it today explain it over the phone to someone who’s never done it? How many times does “it depends” come up?
If it comes up more than twice, the task is more complex than it looks. That doesn’t make automation impossible, but those “it depends” parts must first be turned into rules. That’s often the most valuable part of the work, because once the process is written down the team benefits even without automation.
3. Does the input come from a defined place and the output go to a defined place?
Automation is about taking something from one place to another. If it’s unclear where the input comes from, for example sometimes by email and sometimes verbally, there’s nothing for the automation to hold on to.
In that case the first step isn’t automation but channelling the input into one place. That alone is a significant improvement, and it makes the later automation possible.
4. Does the task require judgement?
A choice that can be expressed as a rule isn’t a judgement call. “If the amount exceeds a threshold, send it to a manager” is a rule and can be automated. “Should we be flexible, given where we are with this client?” is a judgement and should stay with a person.
Flows built without making this distinction clear jam at the first exception. Listing the exceptions up front and deciding where each one goes is the most tedious but most decisive part of the build.
5. If it goes wrong, can the mistake be reversed?
This is the question most often skipped. What happens when a flow runs incorrectly? A wrong report can be corrected. A wrong invoice sent to a client makes things harder. A wrong payment is much more expensive to undo.
Tasks with a high cost of error can still be automated, but a human approval step must always sit in the middle. Adding that step slows the flow by a few minutes; skipping it creates months of mistrust.
Running the filter on an example
In a team I worked with, the task at the top of the list was the “monthly client report”. Everyone complained about it, and the automation conversation started there.
When we applied the filter, the picture changed. The first question wasn’t an issue; the report was repeated every month. We got stuck on the second: the report was prepared slightly differently for each client, and nobody had managed to write down those differences. The third question revealed that the input came from four different places, two regular and two not.
In the same conversation another task came up that was on nobody’s agenda: sorting incoming requests by hand and forwarding them to the right person. It happened several times a day, its steps could be written down clearly, the input came through a single channel, it was rule-based, and a mistake could be reversed. It passed all five questions.
The pilot was built on that second task. The monthly report went into the “standardise first” box, because it needed a decision before it needed automation: why is the format different for each client? Once that question was answered, automation became easy.
The lesson: the filter doesn’t just help you choose the right task, it also shows why the wrong task is wrong. Had we forced the monthly report, the build would have worked but the accuracy of the output would have been argued about constantly.
How to rank the tasks that pass
The filter gives you a list of candidates, not an order. Prioritising automation is a separate decision, and I make it on two axes: how much time the task consumes in a year, and how hard it is to build.
| Annual time burden | Build difficulty | Decision |
|---|---|---|
| High | Low | Do it first. This is the ideal pilot candidate. |
| High | High | Second. Let the pilot succeed and build trust first. |
| Low | Low | Do it when there’s time. Small wins that lift morale. |
| Low | High | Don’t do it. Automating it won’t pay back. |
The last row removes the most tasks from the list, and it’s usually the most resisted decision. If a task is hard and annoying, solving it is tempting. But if its time burden is low, the solution won’t pay you back.
For the pilot, I recommend choosing a single task from the first row. Trying to build three flows at once is the most common way to leave all three half-finished.
Measure the time burden, don’t estimate it
The first column of that table is where most mistakes are made. People are systematically wrong when estimating how long their own tasks take, in both directions: they think the work they hate takes long and the work they’re used to takes little time.
A simple two-week measurement fixes this. Each person writes the date and duration in one shared table every time they do the task. No fancy tool needed; a shared spreadsheet is enough. The picture after two weeks usually surprises everyone.
That measurement has a second benefit: when you repeat it after the build, you can show the gain without any debate. If you didn’t measure first, you can’t defend it later. Enter the measured time into the calculator to see what it adds up to per week, month and year.
What about the tasks that don’t pass?
A task not being an automation-ready process today doesn’t mean it stays manual forever. Tasks that fail the filter don’t go in the bin; they go in a different box. In my experience three boxes emerge.
- Standardise first. Tasks that fail the second and third questions. They’re reassessed once the process is written down.
- Partly automate. Tasks that fail the fourth question. Usually the preparation part is automated and the decision stays with a person. This is often the most efficient model.
- Keep manual. Rare, complex and high-risk tasks. Not automating these isn’t a failure; it’s the right decision.
This is exactly the invisible part of consulting work. Deciding what to automate is easy; deciding what not to automate is hard.
Frequently asked questions
We’re a small team. Is it too early for us?
What matters is repetition, not size. Automating repetitive tasks can make sense in a three-person team too; they may be spending eight hours a week merging reports, and those eight hours are worth more to that team than to a large company. Apply the filter; if you have a candidate, you’re big enough.
Who should apply this filter?
It should be applied together with the person doing the work by hand today. They know the process best and remember the exceptions best. An assessment made only at management level misses the exceptions on the ground.
What if no task passes the filter?
That’s a good result, not a bad one. Either your processes haven’t settled enough, or your repetition burden really is low. In the first case the job is to write the processes down; in the second, to take automation off the agenda for now.
Does using AI change this filter?
It stretches the fourth question a little. Language models can handle some tasks that couldn’t be reduced to rules before, such as classifying the topic of an incoming request. But it doesn’t stretch the fifth. Because a model can be confidently wrong, the human approval requirement becomes even more important for tasks with a high cost of error.
Drawing up your own list
You can apply this filter with your team in an hour. If more than one candidate comes out of it, I recommend starting with a single workflow automation. And if you’d like to take the first step on your own, the five questions above are a good enough start.
Let’s choose the right pilot together
How a build works for tasks that pass the filter, what I deliver and the most common questions are all on the Business Process Automation page.
Updated:


