How to Run a Two-Week AI Pilot: A Step-by-Step Plan

Before you start

• Pick one workflow and one small group of people who do it every week.
• Measure how long the task takes today, by timing it rather than guessing.
• Decide in advance what would count as success, such as a time saved per task or a faster reply to customers.
• Agree what data can and can't go into the AI tool.

Week one: run it in the background

Run the AI step next to the existing process. Nobody relies on it yet. Compare its output with what people actually did, note where it struggles, and adjust your instructions and examples.

Expect the first few days to be rough. Most of the improvement comes from better instructions and clearer examples, not from switching tools.

Week two: use it for real

Let the group use it on real work with a person checking every output. Keep a simple log of each result: accepted as is, edited, or thrown away, and how long the whole task took.

That log will tell you more than any demo. If most outputs need heavy editing, the task or the instructions need rethinking.

Make a clear decision

At the end, choose one of three options: keep going, change the approach, or stop. Base it on the measure you agreed at the start and on what the people using it say.

Stopping is a good result if it saved you from a bigger, more expensive mistake. Keeping going means planning how it will be supported, who owns it, and which workflow comes next.

A simple pilot checklist

• One workflow, one group, one measure.
• A baseline measured before you start.
• A person reviewing every output.
• A log of accepted, edited, and rejected results.
• A decision meeting on a date set in advance.