What to automate first: a method that takes one evening
How to pick the process whose automation pays back, and tell it apart from the one that merely annoys you. Four columns and the ten-minute rule.
The most common mistake in a first automation project is starting with whatever hurts loudest. What hurts loudest is usually whatever irritates the founder personally: an awkward spreadsheet, a broken report, a salesperson who forgets to call back. These are real problems, and they are almost always economically small.
The money sits somewhere else — in work that irritates nobody, because everyone got used to it years ago.
The ten-minute rule
Take any task your team performs and ask two questions:
- How many times a week does someone do it?
- Does it take roughly the same time every time?
If the answer to the second is yes and the answer to the first is more than twenty, you have a candidate. Constant duration means there are no decisions inside the task: the person is not thinking, they are executing a procedure. A procedure can be written as code.
A task that takes five minutes one time and an hour the next is built differently. There is judgement inside it, and you cannot automate it whole. You can automate the preparation — gather the data, lay it out — and leave the decision to a person.
Four columns
Write down five to seven repeating tasks and fill in:
| Task | Times per week | Minutes each | Constant? |
|---|---|---|---|
| Moving enquiries from email to CRM | 60 | 4 | yes |
| Answering “is this in stock?” | 90 | 2 | yes |
| Assembling the weekly report | 1 | 180 | yes |
| Approving a non-standard discount | 8 | 25 | no |
Multiply frequency by minutes. In the example: enquiries 240 minutes a week, stock questions 180, the report 180, discounts 200.
All four look equal. But the third column eliminates discounts: someone thinks every time, and a robot will not think for your sales manager. Three tasks remain, roughly three hours a week each.
Then choose by risk, not by money
Three tasks of equal value automate very differently.
The report is the safest candidate. The data already exists, the result is verified by eye in a minute, and a mistake breaks nothing. Things like this ship in a week and immediately produce a visible result you can show the team.
Moving enquiries carries medium risk. If the robot drops an enquiry, you lose a customer and find out late. So it needs a control: a counter of processed messages reconciled against the mailbox, and an alert when the numbers disagree.
Stock answers carry the highest risk, because here the robot talks to customers in your name. It can answer confidently and wrongly. This ships last, with guard rails: the bot answers only from your system and says “let me check” about everything else.
The right order runs from safe to risky. Not out of caution, but because the first automation has to earn the team’s trust. If you start with a bot that misinforms a customer in week one, you will not get approval for a second project.
What the table does not show
Three things that never make it into the arithmetic and often decide the outcome:
Work nobody does at all. Nobody counts the hours spent on “call everyone who did not buy last month”, because nobody does it — there is no one left. Automation here does not save hours, it creates revenue out of nothing. Keep a separate list of these.
Nights and weekends. An enquiry arriving Friday at 9 p.m. is handled Monday at 10 a.m. Sixty-one hours. In that window the customer found three other suppliers. Automation here does not shorten anyone’s day — it changes who wins the deal.
The cost of an error, not of the time. Someone copies prices into quotes by hand and gets one wrong by a factor of ten every quarter. It takes ten minutes and costs one deal, or one apology visit.
What to do with the result
Take the top row among the safe candidates and describe it in words: what arrives at the input, what should come out, where time leaks today. One page.
With that page you can talk to a supplier, and the conversation will be concrete, because you already know the value at stake and will not be sold “automate everything” in one go.
Often at this point it turns out the task is solved by configuring something you already pay for: your CRM has a built-in workflow, your mail client has rules, your ERP has an exchange format. That is a good outcome and a reason not to spend money.
If it is not solved that way, you are holding a specification rather than a vague feeling that something ought to be done.