How do you find the tasks worth automating?

Ask the people who do the work, not the people who own the business. The owner knows the process as it was designed; the person doing it knows the four extra steps it actually takes, and those extra steps are where the hours are.

Keep a list for one week. It does not need to be precise, and it should include the jobs nobody thinks of as work because they have always been done that way.

  • What the task is, in one line, as the person doing it would describe it.
  • Who does it, and how many times a day or a week it happens.
  • How long one instance takes, including the waiting and the looking things up.
  • Where the information comes from and where it has to end up. Two systems that do not talk to each other is the strongest signal on this list.
  • What happens when it is done wrong, and who finds out.

How do you score the list?

Four questions, each answered high, medium or low. The point is not a precise number, it is an order you can defend when somebody asks why that job went first.

The fourth column is the one people skip, and it is the one that decides whether an automation is safe to run without a person watching. A task that is frequent, slow and checkable but catastrophic when wrong needs a human approval step, which changes the build.

QuestionHigh score looks likeLow score looks likeWhy it matters
How often does it run?Many times a dayOnce a quarterFrequency is what turns a small saving into a real one
How long does one instance take?Minutes of copying and looking upA few secondsTime per instance is the other half of the saving
Can a person check the answer in under a minute?Yes, against a record that already existsNo, it needs judgement or researchIf nobody can check it, the work moved rather than disappeared
What does a mistake cost?An internal correctionA customer, a payment or a legal positionThis decides whether a human stays in the loop

What does a scored list look like?

An illustration with made-up scores, not a client result. The tasks are the kinds that come up in almost every small business, and the ordering is the part worth copying.

Notice that the top line is not the most impressive one. It is the one where the information already exists in two places and somebody is the connector.

TaskOftenSlowCheckableMistake costsVerdict
Copying order details into the accounts packageHighHighYesInternal correctionAutomate first
Answering the same five questions by emailHighMediumYesA wrong answer to a customerAutomate with a draft a person approves
Chasing a missing document before a job can startMediumHighYesA delayAutomate the reminder, keep the judgement
Deciding which supplier to use for a one-off purchaseLowHighNoReal moneyLeave it alone
Writing the monthly update nobody readsLowMediumNoNothingStop doing it instead

What should you leave alone?

Saying no is most of the value of this exercise. These are the patterns I turn down or talk people out of, and the reason is always the same: the automation would cost more than the task.

  • Anything that runs rarely. A quarterly job saved by ten minutes is not a project, however tedious it feels.
  • Anything whose output nobody can check. If verifying the result takes as long as doing it, you have bought a second job.
  • A process that changes every week. Fix the process first, then automate the version that stopped moving.
  • A decision with real consequences and no clear rule. Sorting and drafting are automatable; choosing a supplier or approving a refund is not, unless you can write the rule down.
  • The task you find most interesting. It is rarely the one that pays, and starting there is how a first automation becomes a shelf ornament.

When should you hire someone, and when should you not?

The top line of your list decides this, not the size of your business.

  • Do it yourself when the top line is two systems with clean connectors and a path you can draw. That is an afternoon with a tool like n8n, and it teaches you what the process really is.
  • Hire someone when the top line crosses systems that do not talk cleanly, when a failure costs a customer or a payment, or when a model has to read something unstructured and the result matters.
  • Hire someone when the list is long and you cannot tell which line is first. A short paid piece of work that produces the scored list and a written plan is cheaper than automating the wrong thing.
  • Do not hire anyone to automate everything at once. One task, one owner, one measurable outcome, then the next line on the list.

What do you do with the top line?

Turn it into a small, finishable job rather than a programme. This is the order I run every engagement in, and it is the same whether the answer is a plain automation or an AI step inside one.

  • Write what done means for that one task, including what happens when a step fails and who finds out.
  • Agree one fixed price for it. Small jobs start at $49. A whole process is quoted after a free call, as one price in writing.
  • Build it in your own accounts, so the workflow and the credentials stay with you.
  • Run it alongside the manual version for a week and compare. That week is the evidence that it works, and it is free.
  • Only then take the next line on the list. The score will have changed by then, because the first automation removes work from the others.

What to remember

  • The list is the project. One week of who does what, how often, for how long, and where the information comes from beats any tool comparison.
  • Score on frequency, time per instance, checkability and the cost of a mistake. The fourth decides whether a person stays in the loop.
  • The best first automation is usually boring: moving information between two systems, where you can prove it worked.
  • Start with one task from $49, run it beside the manual version for a week, then take the next line.