Most small companies do not lack automation ideas. They have a list of twenty annoyances and no way to pick between them, so the loudest complaint wins. That is usually not the task that costs the most.
Here is a scoring method you can run in an afternoon with a spreadsheet and the people who do the work.
#Collect the list from the people doing the work
Do not start in a meeting room with managers. Ask every person on the team the same three questions:
- What do you copy, paste or retype from one place to another?
- What do you do on a schedule that someone could check instead of you?
- What do you fix by hand because the data arrived wrong?
Ask for specifics: the task, who does it, how often, and how long one round takes. If someone says “reporting,” ask which report, from which systems, for whom. Vague items cannot be scored.
You will get duplicates. Merge them. You will also get tasks people are embarrassed by (“I retype this PDF into a spreadsheet”). Those are often the best finds.
#Score each task on four factors
Rate each task from 1 to 5 on each of these:
- Frequency. 1 is a few times a year, 5 is many times a day.
- Time per task. 1 is under a minute, 5 is an hour or more.
- Error cost. What happens when it goes wrong? 1 is a shrug, 5 is a lost customer, a wrong invoice or a compliance problem.
- How rule-based it is. 1 means it needs judgment every time, 5 means a new employee could follow a written checklist.
Multiply the four numbers. The highest scores go to work that is frequent, slow, costly when wrong, and predictable. Rule-based is the factor people forget, and it matters most: software is good at following rules and bad at guessing what you meant.
#A worked example (hypothetical)
Imagine a 20-person distribution company. Its team hands in four tasks, and the owner scores them:
- Typing emailed purchase orders into the accounting system: frequency 5, time 3, error cost 4, rule-based 4. Score: 240.
- Building the monthly sales report from three spreadsheets: 2, 4, 3, 5. Score: 120.
- Answering questions about custom quotes: 4, 3, 5, 1. Score: 60.
- Creating accounts for new hires: 1, 3, 2, 5. Score: 30.
The purchase orders go first. They happen all day, a typo becomes a wrong shipment, and the format is predictable. The report is a solid second. Custom quotes score low because every one needs judgment, even though they feel urgent. New-hire accounts are rule-based but rare, so the payoff is small.
These numbers are made up to show the method. Use your own.
#Start with boring internal work
Your first project should be internal, repetitive and low drama: moving data between systems, sending reminders, assembling reports, checking that two records match. Three reasons:
- The users are your own team. They will tell you quickly when something is off, and nobody outside the company sees a mistake.
- The rules are already known. Someone on your team can explain exactly what they do today.
- A win builds trust. A team that sees one tedious chore disappear will hand you better ideas for the second project.
Customer-facing automation can be worth doing, but do it after you have a feel for how these systems behave. If an AI agent is on your list, where AI actually pays off in a small business covers how to tell the real opportunities from the shiny ones.
#What to leave manual
Some work should stay with people, even if it scores moderately:
- Tasks that need judgment or a relationship, like negotiating a price or calming down an upset customer.
- Tasks that change often. If the process is rewritten every quarter, you will pay to rebuild the automation each time. Stabilize the process first.
- Rare tasks with high setup cost. A chore done twice a year rarely justifies building anything.
- Anything nobody understands yet. Automating a confusing process just makes the confusion faster.
A good compromise for gray areas is partial automation: let the system prepare the draft, pull the data or flag the exceptions, and let a person approve the result.
#Turn the top score into a plan
Take the highest-scoring task and write one page: what triggers it, what the inputs are, what the correct output looks like, which cases are exceptions, and who handles those. Our process mapping sheet gives you a format for that.
If you would like a second opinion on your list, you can book a 30-minute call with an engineer. Bring the scored list and we will tell you what we would build first, and what we would leave alone.
