Which tasks should you automate first?
Lees dit in het NederlandsAutomate first the tasks that are repetitive, rule-based and expensive in hours. Repetitive means it comes back on a rhythm you can predict. Rule-based means you can write the steps down without the word "depends" showing up. Expensive in hours means the minutes add up to real time across a year. The way to find those tasks is not a list of tools, it is your own week, checked against six gaps where work leaks: retyping, waiting, gathering, repeating, checking and reporting. Write down every task that sits in one of those gaps, work out what each one costs you in hours a year, and start with the most expensive one.
That is the diagnosis I run before I build anything, and it costs attention rather than a software subscription. The whole method is below, so you can run it without me. If you would rather run it with somebody, it is where I start when I help an owner buy back the hours they lose to repeat work.
Start from your week, not from a tool
The answers I find to this question are lists: tools to buy, or tasks other businesses automated. Neither knows which task is expensive in your week, and a tool can only show you the work it already handles, so you end up automating what the software is good at instead of what your week is bad at.
Take one pass through your calendar, your sent mail and the tabs you keep open, and write down every task you touched twice or more. Do not filter, and put down the boring ones especially: repetition is what makes a task invisible. Then hold each item against the six gaps.
The six gaps
These are the six places I look. They are shapes of work rather than categories of software, so they hold up whatever you build with.
1. Retyping
Data a person carries from one system into another. An order arrives in the shop and gets typed into the bookkeeping tool, or a form submission gets copied into a sheet. The tell is two windows side by side and your eyes going back and forth.
2. Waiting
Work that is finished but has not moved, because it needs a person to pass it on. The invoice is ready and sitting there. The tell is a status that is really a person's name: the thing is "with" somebody, and nothing is produced in the meantime.
3. Gathering
Collecting information from several places before the real work can begin. Before you answer a customer you open the order, the shipping page and the mail thread. The tell is the sentence "let me pull that up."
4. Repeating
The same message written again. The delivery answer, the intake mail, the reminder. The tell is that you go digging in your sent folder for something you already wrote so you can paste it.
5. Checking
A manual look that confirms something a rule could confirm. Did that payment land. Is the stock count still right. Checking feels like responsibility, but a check with one right answer is a rule, and a rule watches continuously instead of whenever you remember.
6. Reporting
The same list or overview assembled every week out of the same sources. The Monday numbers, the open jobs, the what-shipped summary. The tell is that the work is all assembly, because the thinking happens afterward while you read it.
A task can sit in two gaps, and that is a good sign: an order retyped into the books and then rebuilt in Monday's overview means one piece of plumbing pays twice.
Rank them with the cost line
The list is not the answer yet, because the loudest task is not always the most expensive one. Sort it with arithmetic.
hours a year = times per week x minutes each time x 52 / 60
cost a year = hours a year x what an hour of that person's time costs youA Gregorian calendar year is 52 weeks plus a day, or two in a leap year, so 52 is the honest multiplier for a weekly task. Use 12 for a monthly one. Count the whole task including the switching, and use the rate of whoever actually does it, which is not always yours.
The numbers below are invented, made up to show the shape of the calculation.
| Task (example only) | Times a week | Minutes each time | Hours a year |
|---|---|---|---|
| Retyping orders into the bookkeeping tool | 12 | 4 | 42 |
| Answering the same "where is it" mail | 6 | 5 | 26 |
| Building the Monday overview | 1 | 45 | 39 |
In that example the Monday overview feels like a small weekly chore and costs almost as much as the retyping done a couple of times a day. Do the arithmetic instead of trusting the feeling, then multiply each row by what an hour of that person's time costs to get your order of work.
What not to automate first
Some work is a poor first candidate even when it repeats.
- Judgment-heavy work. If the right answer depends on reading a situation, you are not automating a task, you are automating a decision.
- Rare work. The annual filing hurts and it happens once. Painful is not the criterion, cost a year is.
- Work that changes every time. If the steps differed on the last three occasions there is no rule yet. Write the rule first, then automate it.
- Anything where a wrong output costs more than the hours it saves. Money leaving the account, a message going to a customer, anything with a legal or medical edge. The saving is bounded, the mistake is not.
That last one is not an argument against automating those jobs. It is an argument about where the person stands.
Automate the steps, keep the person at the decision
Nearly every task worth automating splits into a deterministic part and a decision. The deterministic part is fetching, formatting, calculating, moving and preparing. The decision is press send, pay this, tell the customer no. Hand over the first part completely and keep a person on the second.
In practice: draft, do not send. Prepare the payment, do not pay it. Score the thing, do not judge it. The Answer Report, which is my own product, is built that way. It measures, publishes the score and hands over the fix, and every check is deterministic with each result quoting the exact page text behind it. The measuring is the machine's part. The deciding is not.
I did the same on my own webshop. The order flow and the customer mail went first, because both were textbook cases: an order landing in one system and needing to exist in another is retyping, and the same status message written again for every order is repeating. I spent years in kitchens before this, where the prep list did the same job: the line ran whether or not the chef was standing on it.
Once it runs on its own, give it a way to tell you when it fails and read its first stretch of output by hand, because an automation that breaks quietly is worse than the task it replaced.
How to run this on your own week
- Write down every task you touched twice or more this week, unfiltered.
- Tag each one with a gap: retyping, waiting, gathering, repeating, checking, reporting. Anything that takes no tag comes off the list.
- Run the cost line on what is left and sort by hours a year.
- Split the top row in two, the deterministic steps and the decision at the end.
- Automate the steps, keep the decision with a person, and give the thing an alarm.
Then take the next row. The order matters more than the tooling: a plain automation on your most expensive task beats a clever one on your third. The projects on my work page each began with a list like this one. If you want a second pair of eyes on which row to take first, that is what a conversation about your first automation is for.
Sources
Frequently asked questions
How do I know whether a task is rule-based enough to automate?
Write the task down as instructions somebody new could follow without asking you a single question. If that takes a few lines and the steps do not shift with the situation, it is rule-based. If the instructions keep needing "it depends", the task carries judgment and belongs with a person for now, or belongs to a person at the decision point while software does the steps around it.
What if a task only takes a minute?
Work out its yearly cost before you dismiss it: how often it happens, times the minutes it takes, across a year. A one-minute job done many times a day can cost more hours in a year than a long job you do once a month, and the switching around a short task is often bigger than the task. Length is a feeling, hours a year is a number.
Should I automate the task I hate the most?
Not automatically. Annoyance is a useful signal that something is repetitive, but the job you hate may be rare, and a rare job cannot cost you much time however unpleasant it is. Put it through the same calculation as everything else, hours a year times what an hour of that time costs, and let the number decide the order.
What do I do about the tasks that need judgment?
Split them instead of skipping them. Automate the fetching, formatting and preparing, then stop at the point where somebody has to choose, and let the software present the choice with everything needed to make it. You keep the decision and still lose most of the manual work around it.
Can I run this diagnosis if I work alone?
Yes. The six gaps, retyping, waiting, gathering, repeating, checking and reporting, do not change when there is nobody to hand work to. Waiting simply looks different: instead of a job sitting with a colleague it sits with you until you get to it, which is still a delay with no work happening inside it. Use your own hourly cost in the calculation, since with no staff your time is the only rate there is.