← All guides

FIELD GUIDE / TEST THE CHANGE

Run a small AI pilot that can answer a real question.

A simple scorecard for comparing a workflow before and after one change.

Choose one outcome and one owner.

Pick a repeatable task and a question the team can answer, such as: “Can we prepare accurate inquiry drafts with less review time?” Name the person responsible for the trial, its start and end, the permitted inputs and the human who reviews outputs. Keep high-consequence decisions and external actions with the authorized person.

Before the trial, agree what counts as completed correctly, what counts as a correction, and what would make you pause. Use the same definitions and similar types of work in both periods. Small samples can be noisy; keep individual examples alongside the totals.

Compare two periods.

Enter counts for comparable work and the total minutes people spent reviewing and correcting it in each period. Blank fields are incomplete, not zero. Entries stay on this page and are not sent or stored.

Before
During pilot
Enter all six values to compare the two periods.

Completion rate = correctly completed ÷ reviewed. Review time is minutes per reviewed item, including corrections. This is not a savings or ROI forecast.

Look behind the averages.

Review a few successes and failures. Did the same errors recur? Was an item marked complete but sent to the wrong place? Did someone spend extra time chasing an exception outside the tool? Record the time and work the scorecard misses, including setup and maintenance.

For example, 8 correct of 10 with 50 review minutes is 80% and 5 minutes per item; 9 of 10 with 30 minutes is 90% and 3 minutes. That is useful evidence for a next conversation, not proof of a lasting improvement.

Decide: continue, adjust or stop.

Continue only if the agreed outcome and review process are working. Adjust one failure point and test again if the results are mixed. Stop or return to the prior process if errors or unresolved exceptions exceed your agreed limit. Name an owner for monitoring and a date to revisit the decision.

Use the workflow diagnostic to investigate a recurring failure, and the value guide to separate cash savings from time freed up. For broader evaluation and oversight considerations, see NIST’s AI evaluation planning manual.

Examples are illustrative, not customer results.

Meet the team →

LESS OVERLAP. MORE PURPOSE.

Make your next move a clear one.

Start with a view of the tools you use. Then decide what deserves a closer look.

A CLEANER SYSTEM STARTS HERE

Let’s untangle
your AI.

Tell us what you’ve tried.
We’ll start with what needs to work better.

A few lines are plenty.

Share your audit Optional

We’ll receive the tools, costs and answers shown below to prepare for your conversation. This does not subscribe you to marketing.

We’ll use these details to respond
to your inquiry.