The short answer
Start with a frequent, predictable task that is easy to check and safe to stop. Automation (having software carry out repeatable steps) is a better first experiment when the information is reliable and someone can handle mistakes. The task taking the most time is not automatically the best starting point.
Pick three tasks, not three software products
Ask the people doing the work: “What do you copy, check or chase repeatedly?” Name a specific task with a clear finish. “Prepare the weekly status report” is easier to assess than “automate the office.”
For each task, write down how often it happens, the hands-on minutes it takes and the corrections it needs. Keep waiting time separate from working time.
Example — not a client result
A task done 30 times a week at 5 minutes each uses 150 minutes. That is the current effort, not a promised saving. Any review, repairs and ongoing care still take time after automation.
Pause if any of these are unresolved
- The rules keep changing. Agree how the task should work before repeating it with software.
- The information is unreliable. Fix missing or inconsistent details before building around them.
- A mistake would be hard to undo. Keep a person approving payments, customer messages or important record changes until checks and recovery steps are agreed.
- No one will own it. Name the person who checks failures and can stop the process.
A high worksheet score does not override these concerns. Resolve them or choose a smaller task.
Compare your shortlist on four questions
Give each task 0, 1 or 2 points per question. This is a simple discussion tool, not a tested prediction of savings or project success. Use the same meaning of “often” and “useful” for all three tasks.
| Question | 0 points | 1 point | 2 points |
|---|---|---|---|
| Does it happen often? | Rarely | Some weeks | Most working days |
| Are the steps predictable? | Different each time | Some usual steps | Clear steps, few unusual cases |
| Is the information ready? | Missing or inconsistent | Some cleanup needed | Complete and consistent |
| Would less manual work help? | Little practical benefit | Helpful improvement | Frees time for a named priority |
Add the four scores: up to 8 points. A higher total gives you a task to investigate first. It does not mean you should buy software or approve a build.
What a shortlist might look like
These made-up examples show how to reason about a choice. Your tools, volume and rules may give a different answer.
Prepare an internal report
If staff copy the same approved figures each day, a small trial may be easy to check against the original records.
First check: can the current reporting tool already do it?
Draft customer replies
Preparing a draft may help, but a person should check names, facts and promises before it is sent.
First check: can you separate drafting from sending?
Handle unusual refunds
If each case needs a different judgment, first document who decides and what they need to know.
First check: is a clear review checklist enough?
Keep this worksheet
Compare three tasks with your team
Download the plain-text worksheet and fill it in using your own notes. No email address is required.
- Name each task: what starts it and what counts as finished?
- Record the effort: times per week × hands-on minutes, plus correction time not already counted.
- Score the four questions: frequency, clear steps, ready information and practical benefit.
- Check the pause conditions: rules, information, consequences and owner.
- Estimate the extra work: setup, checking results, fixing failures and ongoing costs.
- Choose one small trial: name the owner and write down what would make you continue, change or stop.
Try the smallest useful improvement
Check your existing software before replacing it. A saved report, a better form or a simple connection between tools may be enough. See how to compare existing software with a custom app.
Try the chosen task on a small set of examples, with a person checking the results. Compare the total work before and after, including repairs. Use the automation preparation checklist to plan the test.
Common questions
What if two tasks get the same score?
Start with the one that is easier to check, cheaper to try and safer to stop. If neither has a clear owner, pause both.
Do we need AI for the first project?
Not necessarily. AI (artificial intelligence) may help interpret varied information, but a predictable copying task may only need fixed rules. Choose the task before choosing the technology.
What if none of our tasks looks ready?
That is a useful finding. Improve the instructions or information first, then reassess. Keeping a task manual can be the right decision.