YOUR ONE-PAGE CASE

PILOT PROPOSAL: [what the AI will do]
From: [name and role] · For: [approver]

THE PROBLEM
[problem]. It affects [who]. Today it happens about [volume] times a month and takes about [minutes] minutes each: roughly [H] hours a month. ([measured / a good estimate / a rough guess].) What goes wrong: [what goes wrong].

THE PROPOSAL
For [weeks], [N] people will use [tool] to [in a sentence]. People will still [what people still do]. It will use [information] only.
Out of scope: Anything about an individual's own case; Sending anything without a person checking it.

HOW WE'LL KNOW IT WORKS
- Time: from [minutes] to [minutes with AI] minutes per item, including checking. Across the whole task, that's about [S] hours a month.
- Quality: A reviewer samples 20 outputs a week and counts how many needed a factual correction. Target: [target].
- People: [feelings measure].
We stop if: A mistake reaches someone outside the pilot; Quality stays below target for two weeks in a row.

RISKS AND CONTROLS
- Personal or confidential information: [status]. [control or note]
- Wrong or invented answers: [status]. [control or note]
- Approved tool: [status]. [control or note]
- Who's affected: [status]. [control or note]
- Being open about it: [status]. [control or note]
- Security: [status]. [control or note]
- Accountability: [status]. [control or note]
- Fallback: [status]. [control or note]

WHAT I'M ASKING FOR
Approval to run a [weeks] pilot with [N] people using [tool].

MEASUREMENT PLAN

RISKS AND CONTROLS

PILOT PLAN

Week 1: Brief the pilot users, measure the baseline, set up the tool and sources.

Final week: Review the results against the targets and decide: keep, change or stop.

APPROVER QUESTIONS AND ANSWERS

What if it gets something wrong?
[what people still do]. We stop if A mistake reaches someone outside the pilot.

Is our data safe?
[personal or confidential information control]. [approved tool control].

How much time will it really save?
We estimate about [S] hours a month across the task. The pilot measures it with [N] people over [weeks].

What does it cost?
It uses [tool], which we already have. The main cost is the pilot users' time to check and record results.

Why this, and why now?
[what goes wrong].

STILL TO DO

• □What's the problem, in one sentence?
• □Who has the problem?
• □What's your idea for AI?
• □Question 1 in the idea check
• □Question 2 in the idea check
• □Question 3 in the idea check
• □Question 4 in the idea check
• □How many times does this happen a month?
• □How many minutes does each one take today?
• □How many people do this work?
• □What goes wrong or takes longest?
• □How do you know these numbers?
• □What will the AI do?
• □What the AI will do, in a sentence
• □What will people still do?
• □Who's in the pilot?
• □How long?
• □Which AI tool?
• □What information will it use?
• □Minutes per item with AI, including checking
• □Quality target
• □How will the people using it feel?
• □Personal or confidential information: needs an answer
• □Wrong or invented answers: needs an answer
• □Approved tool: needs an answer
• □Who's affected: needs an answer
• □Being open about it: needs an answer
• □Security: needs an answer
• □Accountability: needs an answer
• □Fallback: needs an answer
• □Your name and role
• □Who will approve it?
• □What are you asking for?
• □Decision