AI PATH PRO / TEXT EDITION
Read it and put it to work.
AI Path Pro · 45 minutes + measurement
PILOT CANVAS
Leave with a pilot plan: task, hypothesis, comparison, quality gate and decision owner. Use it for a process improvement or an AI solution.
Try. Check. Decide.
01 / DESIGN THE DECISION
A useful trial answers a question.
Avoid changing an entire operation at once. Choose a recurring task, one bounded change and the decision you want to make at the end. Before starting, define what would justify continuing, changing direction or stopping.
Define the eligible cases.
Name the task, participants, starting state and completed state. Specify which exceptions stay outside this trial, and keep their share visible. A narrow scope is useful only when its limits are understood.
Measure the current work.
For comparable cases, record total working time, quality and waiting. Include preparation, review and rework. Label estimates explicitly; measured values should have a traceable log. Keep the unit and the definition of each measure consistent.
Make the hypothesis testable.
Use this structure: changing this one thing will improve this measure while preserving this quality condition. The target is a decision threshold, not an advance announcement of the result. State which observations would challenge the hypothesis.
Test the awkward cases.
Include normal, incomplete, conflicting, repeated and interrupted tasks. Write the expected behaviour first. Give exceptions an owner and the new process a rollback path. Check recovery as well as the first successful run.
Review performance and actual use.
Compare like-for-like cases and examine user feedback. Show implementation, training and maintenance effort separately. Decide to continue, adjust or stop based on the evidence. Record the next action and its owner.
02 / WORKED EXAMPLE
One intake form, ten new cases.
A fictional learning example. Baseline figures are illustrative data, and targets are planning assumptions. Ten comparable routine requests are followed by ten new eligible cases. The trial has not yet produced a result.
| Measure | Illustrative baseline | Target chosen in advance |
|---|---|---|
| Total working time | Mean across 10 cases: 32 minutes/case. | No more than 24 minutes/case on average. |
| Review and rework | 8 minutes/case within the total of 32. | No more than 8 minutes/case; included in total time. |
| Accepted first time | 8 of the 10 cases. | At least 9 of the 10 cases. |
| Waiting time | Median across the ten cases: 18 hours. | A median of no more than 8 hours. |
| Use of the new form | The new form is not yet in use. | Used for at least eight of the ten eligible new cases. |
Compare the same kind of case.
Keep routine and exceptional requests separate. Record exclusions and case counts. A small sample supports an initial decision; gather broader experience before extending the process more widely.
Count the complete effort.
The 8 minutes of review are already inside the total of 32: do not add them twice. Record creating the form and training users separately as one-off effort.
Read the targets together.
Faster work needs acceptable quality and actual use. If colleagues bypass the form, find out why. A solution that is difficult to use needs another round of improvement.
03 / YOUR PILOT
Six fields that support a decision.
Complete this before the trial and review it with the process owner. If targets or stopping conditions change later, record a new decision rather than redefining success after seeing the results.
- 01 / TASK AND BOUNDARY
What will you try, with whom and on which cases? What stays outside the trial?
Example answer: Routine internal requests with two colleagues. Urgent and unusual requests tracked separately.
- 02 / BASELINE AND MEASUREMENT
Which source, period and unit will provide comparable data?
Example answer: Case log: preparation + execution + review + rework, in minutes. Waiting recorded separately, in hours.
- 03 / HYPOTHESIS AND TARGET
What changes, what improvement do you expect, and which quality condition must hold?
Example answer: Common form: mean ≤24 minutes/case; at least 9/10 outputs accepted first time.
- 04 / TESTS AND EXCEPTIONS
What should happen with normal, incomplete, repeated and interrupted cases?
Example answer: Ask for missing fields; create no duplicate case for a repeated ID; retain visible status after an interruption.
- 05 / OWNER AND ROLLBACK
Who supports users, handles failures and decides whether to restore the previous process?
Example answer: Named process owner; shared issue log; documented previous process available if a rollback is needed.
- 06 / DECISION AND NEXT STEP
When will you decide, using which evidence and conditions?
Example answer: After ten eligible new cases: time, quality, use and feedback. Continue, adjust or stop, with a recorded reason.
04 / DECISION GATES
Agree how you will decide.
Adapt these gates to your task. Write down the specific action each shortfall requires: which evidence, change or decision is needed, and who is responsible for it.
- The measurement is usable
Ready: Comparable cases, complete effort and visible exceptions support the review.
Still to do: The samples differ or data is missing. Complete the measurement and mark the current conclusion as uncertain.
- Quality is acceptable
Ready: The predefined quality condition is met, with errors and rework included in the record.
Still to do: Speed came with errors or lost controls. Improve and retest, or restore the previous process.
- People actually use it
Ready: The intended users adopt it for eligible cases and know when to ask for help.
Still to do: People bypass it or exceptions stall. Investigate the cause with the users.
- 15 MINUTES
Agree the hypothesis together.
Ask the process owner and one person doing the work to complete the six fields. If a baseline is missing, arranging that measurement comes first. Distinguish known values from estimates.
- 20 MINUTES
Prepare the trial log.
Give each row a case ID, case type, working time, waiting, quality status, rework and user feedback. Complete one row together to check that each field means the same thing to everyone.
- 10 MINUTES
Book the decision point.
Name the trial lead and the final decision owner. Set a case count or review date, list the evidence required, and prepare for three outcomes: continue, adjust or stop.
Related sources
- Make: Scenarios
For the technical implementation when your trial concerns a workflow controlled by explicit rules.
- v0 documentation
For a trial application built around one clearly bounded user task.
Stopping a weak idea can be a useful result of the pilot.