Every small business has a list of things that feel like they should be automated. The list is usually long, and the temptation is to start with whatever is most annoying this week. That is how you end up with an elaborate automation for something that happens twice a month.
A better approach is to score candidates against a few plain criteria and let the ranking surprise you. You do not need numbers or a spreadsheet model. High, medium and low are enough, as long as you apply them consistently.
Write the candidates as tasks, not tools
Start by listing the actual tasks people repeat, in the words they use. “Chasing late invoices”, “copying web enquiries into the CRM”, “assembling the Monday report”, “setting up a new client folder and sending the welcome email”. Avoid phrasing like “implement a chatbot”, which smuggles in a solution before you know the problem.
Ask the people who do the work. Owners often guess wrong about where time goes, because the painful tasks are not always the frequent ones.
Score on five questions
For each task, rate five things as high, medium or low. The point is comparison, so be honest and quick rather than precise.
- Frequency: how often does this happen? Daily and weekly are high, occasional is low.
- Effort: how much human time does each occurrence take, including interruptions and context switching?
- Rule clarity: can you write the steps as if-this-then-that, or does it depend on judgment and tone?
- Cost of error: if this is done wrong or forgotten, what is the damage? Lost revenue, a compliance issue and an upset customer are high.
- Access: do the systems involved have an integration or API, or is the data trapped in email and PDFs?
Frequency, effort and cost of error tell you the value. Rule clarity and access tell you how hard it will be.
Worked examples
Take four tasks from a typical service business. Chasing overdue invoices is high on frequency, medium on effort, high on cost of error because late payment hurts cash flow, high on rule clarity (a reminder schedule based on days overdue), and high on access because accounting software exposes the data. That is a strong candidate: high value and low difficulty.
Copying web enquiries into the CRM is high on frequency, low to medium on effort per enquiry, high on cost of error because a missed lead is lost revenue, high on rule clarity, and high on access when the form can post to the CRM directly. Another strong candidate, and often the very first one worth doing.
Assembling the Monday report is medium on frequency, high on effort because it involves pulling from several tools, low on cost of error, medium on rule clarity, and medium on access. Worth doing, but second wave, and an exports-to-dashboard fix may beat a custom automation.
Replying to complaints is medium on frequency and high on effort, but it scores low on rule clarity because tone and context matter, and high on cost of error. This is a poor automation candidate. At most, use a draft for a person to edit.
Read the pattern, not the totals
Look for tasks that are high on frequency or cost of error, and also high on rule clarity and access. Those are your first build. Tasks that are valuable but low on rule clarity usually need process work first: agree what the rule is, write it down, and then automate it. Automating a confused process just makes the confusion faster.
Be wary of tasks that are low on access. If the data lives in a system with no integration, the automation cost can exceed the saving. Sometimes the right first step is replacing the tool, not wiring around it.
Start small and keep a person in reach
Build the first automation to handle the simple, common case and send anything unusual to a human. Log what it does so you can see mistakes. Run it alongside the manual process for a short period if the cost of error is high.
Then review after a month. Is it saving the time you expected? Did it create new exceptions? Automation has a maintenance cost, since tools change and rules drift, so give each one an owner. If nobody owns it, it will eventually fail silently.
And accept that some things should stay manual. A task that happens rarely, depends on judgment, and builds a relationship is often better done by a person.
Takeaways
- List real tasks in the words of the people who do them, not tools.
- Score frequency, effort, rule clarity, cost of error and access as high, medium or low.
- First builds are valuable and easy: frequent or costly tasks with clear rules and accessible data.
- Fix an unclear process before automating it.
- Give every automation an owner, a log and a human escape hatch.
Related services
