# AI training pilot guide: five steps to a business decision

A pilot should answer a business question. An AI training pilot guide helps you turn a capability question into a business question: which role, which recurring problem, what baseline, who accepts the result and when you decide. Read it before the first demo, because those answers shape what the demo should show and whether the pilot is worth running at all. The physical therapy practice in the worked example is fictional.

Updated October 26, 2026

## 01 · Fix the question

Choose one role and one recurring problem. Write the outcome in a sentence. Name the person who can accept the result and the date when a decision will be made.

## 02 · Establish a usable baseline

Use comparable recent work where possible. Record cohort size, role, period, task rubric, and support effort. If no reliable baseline exists, measurement is the first pilot activity.

## 03 · Bound the experiment

Agree on source access, environment, outputs, cost cap, review responsibilities, and a stop condition. Do not use production secrets or employee-sensitive records in a demonstration.

## 04 · Observe real users

Ask people to perform the task in the Web workspace without coaching. Record completion, wrong turns, confusing labels, broken layout, and recovery. Retest fixes independently.

## 05 · Decide, with the caveats visible

Compare the agreed metrics with the baseline. Document quality and experience, not only speed. Choose to stop, improve, or expand; do not treat a passed demo as proof of commercial ROI.

## Who should run the pilot, and when?

The person who will actually run the pilot: an HR lead, an operations manager, an enablement manager, or a founder. Read this before the first demo, not after, because the questions it asks shape what the demo should show. If you cannot name the role, the recurring problem, and the person who will accept the result, the pilot is not ready to start, and that is useful to know early.

## A worked example

A multi-site physical therapy practice with seven clinics pilots front-desk intake onboarding. The question: can a new front-desk hire complete a full patient intake without help by the end of week one, with fewer questions to the office manager? The baseline: the last three hires took a median of nine working days and asked about the insurance verification steps most often.

The bound: the Intake Procedure, the Insurance Verification Checklist, and the Front Desk Scripts, with no patient records, reviewed by the practice administrator. The observation: two new hires perform a simulated intake in Web while the office manager watches and records wrong turns. The decision: continue if both pass the intake check by day six with fewer clarifications than the baseline, change if the material needed heavy correction, stop if either fails and the corrections point to the source rather than the lesson.

## What to check before rollout

Five checks before you commit to the pilot.

- One accountable owner can say whether the pilot succeeded or failed.
- The baseline is written down with cohort size and period.
- The stop condition is agreed before the pilot starts.
- The person verifying the result is not the person who built it.
- The decision date is on a calendar.

## Where does a person stay in charge?

Elvina produces drafts, practice, and reports; the pilot owner approves the sources, decides what reaches participants, and makes the continue, change, or stop call. A passed demo shows the workflow can run; it does not show that your team is ready or that the economics hold. Keep both statements separate in the write-up.