# Run a two-week AI trial without making a mess

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

## Useful result

You'll finish with a contained trial plan with a clear stop condition.

A two-week trial should answer a question, not generate a celebratory slide. Decide what you will measure before the team starts experimenting. Keep the trial small enough that ordinary work can continue.

## Set up

Allow 30 minutes to plan, then measure during ten working days. Choose an approved AI tool, one task and one reviewer. The trial duration is not the time needed to read this guide.

## Practice input

Trial: format internal handover notes. Baseline: 15 minutes per handover. Five trial outputs planned. Reviewer: Jo. Exclusions: customer records and external sending. Success: acceptable facts and less total preparation plus review time.

## Do the task

1. Record the existing method and measure one ordinary run.
2. Define sample inputs, review checks and a place to log mistakes.
3. Choose a review date and a stop rule before inviting the team to participate.

## Prompt

Draft a two-week trial checklist for this handover task. Include baseline measurement, five sample runs, factual checks, correction time, reviewer and a final continue/change/stop decision.

## Refine

Add a stopping condition for repeated invented actions or unexpected access requests. Do not promise a percentage saving before the trial runs.

## Expected checks

The checklist should count review and correction time, not just generation speed. Every output needs a named reviewer and no external sending is included.

## Save

Save trial-plan.md and a blank run log. Bring the actual results to the review instead of a best-looking example.

## Sources

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