# Automate a process only after it works manually

Original 5HO practice material. Fictional example. Reviewed 5 October 2026.

## Useful result

You'll finish with a clear test for whether a manual process is ready to automate.

Automation repeats your process. That is excellent when the process works and quite efficient at spreading mistakes when it does not. Run it manually before giving it a schedule.

## Set up

Allow 20 minutes to document the flow, plus several manual trials. Use copied inputs and an output you can inspect without triggering real messages or payments.

## Practice input

Process: read a weekly CSV, total approved expenses, draft a summary. Test file has one rejected claim, one blank approval field and two approved amounts: AUD 20 and AUD 35.

## Do the task

1. Write the filter rule and calculate the expected result manually.
2. Run the task on normal, empty and incomplete inputs.
3. Only after those pass, decide the trigger, log, reviewer and stop mechanism for automation.

## Prompt

Describe a manual test plan for this expense-summary process. Include approved-only filtering, blank approval handling, empty files and a repeated run. Keep the result as an unsent draft.

## Refine

Explain how a repeated run would avoid sending or creating duplicate records if we automate later. Do not claim that a prompt alone enforces that behaviour.

## Expected checks

The approved total is AUD 55. The blank approval goes to review and the rejected claim is excluded. A schedule is a trigger, not proof the process handles errors.

## Save

Save the verified procedure and edge-case results. Treat automation as the next separate implementation task.

## Sources

- Writing clear prompts: https://help.openai.com/en/articles/10032626-prompt-engineering-best-practices-for-chatgpt