A quiet reading corner in a modern office at blue hour, an armchair and floor lamp by tall glazing

The test

How to tell which jobs are worth automating

Two things decide it: how repeatable the job is, and how much judgement it needs. Work that runs the same way every time and requires nobody to decide anything is worth automating, and almost nothing else is, whatever the software can technically do.

Last checked 19 September 2026

The two axes #

Every job in a business sits somewhere on two scales, and where it sits tells you almost everything.

Repeatability

  • Whether the job runs the same way each time
  • A consistent trigger, a consistent sequence and a consistent output
  • Two different people doing it would produce the same result

If the answer to "how do you do this" is "well, it depends", the job is less repeatable than it feels.

Judgement

  • How much of the job is a decision
  • Not a lookup, not a rule, a decision
  • Something where a person weighs context, reads a relationship, applies experience or takes responsibility for being wrong

Judgement is not the same as expertise. A great deal of expert work is actually rule following performed quickly by someone who knows the rules.

What the four quadrants mean in practice #

Low judgement High judgement
High repeatability Automate this. It is the only quadrant that reliably pays Automate the gathering, leave the deciding. Usually the biggest prize
Low repeatability Leave it. The setup costs more than the job Leave it entirely. This is what you employ people for

The top right is where most businesses find their real opportunity, and it is the one that gets missed. A quote that takes an hour is not one task. It is forty minutes of pulling together prices, measurements, previous jobs and availability, and twenty minutes of deciding what to charge this particular customer.

Three questions that split a job properly #

Take any job that eats your week and ask these in order. They are the questions we work through on the free audit, and they work perfectly well on your own.

  1. What starts it?

    If the trigger is a specific event, an email arriving, a form submitted, a date passing, a status changing, that is a strong start. If the trigger is "when someone remembers", the job is not yet repeatable enough to automate, and making it repeatable is the first fix.

  2. Where does the information come from?

    List the places the person has to look. Most repeated admin is repeated because information lives in three systems and a person is the bridge between them. Count the bridges: that number, more than anything, is what makes a job expensive.

  3. What decision is being made, and by whom?

    Be strict here. "Choosing which template to send" is usually a rule wearing a disguise. "Deciding whether to chase this client again this week" is a real judgement, because it depends on the relationship.

Four things that disqualify an otherwise good candidate #

A job can pass both axes and still be the wrong place to start.

  • It is about to change. A process that a new system, a new regulation or a new member of staff is about to reshape should be left alone until it settles. Automating a process twice costs more than automating it late.
  • Nobody actually owns it. If no single person can say how the job is supposed to run, there is nothing to hand over. What gets built is somebody's guess at the process, and it will be wrong in ways nobody can adjudicate.
  • It is rare. Volume is what pays for the work. A job done twice a year can be the most infuriating thing in the business and still not be worth a line of code.
  • The decision has a legal or similarly significant effect on a person. Declining an application, setting a price for an individual, screening a candidate. UK data protection law places specific safeguards on decisions made solely by automated means where the effect on someone is significant, including a right to human involvement. That is not a reason never to touch such a process, but it is a reason not to start there.

Rank by hours, not by irritation #

Once you have candidates, the order is decided by hours, and hours are almost always misremembered. People underestimate jobs they do in small slices and overestimate jobs they dread. The only reliable method is to count occurrences and multiply, rather than to ask somebody how long they think they spend.

Our cost of admin time calculator does that arithmetic and puts a real employer cost on the result, using the current employer National Insurance and pension rates rather than a raw hourly wage. It counts what you spend now. It deliberately does not predict what you would save, because nobody can know that before looking at the job.

The honest conclusion this test often reaches #

Frequently the answer is that a business has two or three genuinely repeatable jobs and a great deal of work that merely feels repetitive, and that the right move is to fix a configuration, change a process or hire someone. A page on the jobs that are not worth automating sets out the patterns where that is the case.

That is not a disappointing outcome. Knowing that the repeated work in your business amounts to four hours a week is worth knowing, because it tells you to stop looking at software and start looking somewhere else.

Questions people ask

What is the quickest test for whether a job is worth automating?
Two questions. How repeatable is it, meaning does it run the same way every time regardless of who does it, and how much judgement does it need, meaning does somebody have to decide something a rule could not decide for them. High repeatability with low judgement is the only combination worth starting with.
How many times a week does a job need to happen to be worth it?
There is no universal threshold, because it depends on how long each occurrence takes and how long the job survives before something changes it. A five minute job done forty times a week is a stronger candidate than a two hour job done once a month, and both are weaker than either looks if the process is about to change.
Should I automate the job that annoys people most?
Only if it is also repeatable. Annoyance is a good signal that something is wrong and a poor signal about what to do. Plenty of deeply irritating work is irritating precisely because it needs judgement under time pressure, and that is the work automation handles worst.
Can a job be half automated?
Almost every worthwhile one is. The usual shape is that software does the gathering, the sorting and the drafting, and a person does the deciding and the sending. Splitting a job in two along the judgement line is more often the answer than automating the whole thing.

Where these numbers come from

  1. ICO, Automated decision-making, including profiling , read 19 September 2026 . Sets out the safeguards that apply where a decision with a legal or similarly significant effect is made solely by automated means.

Last checked 19 September 2026.

Our workings are on the methodology page .

Free audit

Find out what the repeated work costs you

The audit counts every job the business repeats, ranks them by the hours they eat, and maps the five worth automating first. It is free, you keep everything, and there is no obligation at the end of it.

AI automation, explained

Plain answers to what businesses ask before they automate anything.

All of ai automation, explained

Read next