How to choose your first useful automation

Describe the recurring inconvenience

Write the problem as something another person could observe: every week, someone copies the same information into a report and then checks it again. Add who does the work, what they receive and what they need to produce. This gives a discussion about automation a concrete starting point.

Microsoft’s automation guidance recommends considering procedural tasks, their continuing value and the effort required to maintain them. It also recognizes that a workflow can retain human decisions while automating other steps.

List a few candidates from ordinary work. For each, ask someone who performs it to describe the most recent difficult case. That conversation often exposes the real task.

Check whether the inputs are understandable

Take a fictional sample of the information the task receives. Can two people explain each field in the same way? Does an empty value mean zero, unknown or still pending? Resolve those differences before trying to move the information automatically.

Choose an initial boundary you can explain in one sentence. For example, collect completed forms into a draft summary. Keep incomplete forms in a visible review queue. Write down which inputs belong in the pilot and which require a person. The boundary makes it easier to recognize when the automation encounters a case it cannot finish.

Rehearse a small synthetic task

Synthetic example: a fictional training room uses three identical checklists to record chairs, markers and cables. Someone combines them into a weekly readiness report. All materials and situations described here are invented; they do not represent a customer or a measured result.

A first pilot could prepare that report as a draft. Include one complete checklist, one with a missing date and one submitted twice. Decide beforehand what the draft should contain and what should require attention. Add a conflicting count and check that a person can find the disagreement.

This creates a small, inspectable result. It also gives the team concrete examples for explaining what the automation is allowed to do.

Make unfinished work easy to find

An exception needs an owner and a visible next step. In the training-room example, the missing date should lead to a request for clarification. It should remain distinguishable from a completed entry. The person reviewing the draft needs enough context to resolve the issue without starting the whole task again.

Try interrupting the rehearsal and running it again. Check whether a repeated input creates a duplicate result. Agree how to discard a draft and return to the existing manual task. Keep the original test materials available so a confusing result can be compared with what went in.

Compare the complete effort

Record the time spent preparing inputs, reviewing the result and fixing exceptions. Compare it with the same task performed manually on equivalent synthetic material. Note the quality differences too: a faster report that still requires a full recount may offer little practical help.

Ask the person doing the work whether the draft is easier to understand and correct. Include the effort of keeping the automation working when a form changes. Use those observations to choose the next step: continue the pilot, narrow it, improve the inputs or stop. A first automation earns a wider role through evidence from the task it actually handles.