How this page works
The lab steps are in labs.md. This page is where you write your answers for Labs 5 to 8. It works on a phone or a laptop.
- Type your answers into the boxes. They are saved on this device as you type, so refreshing the page does not lose them. Neither does closing it, unless you are in a private (incognito) window. Nothing you type here is sent anywhere, and there is no sign-in.
- Required or optional: boxes marked Required are the ones the lab asks you to fill in today. Boxes marked Optional are for back at work. Each worksheet has a Check for required info button that lists what is still missing.
- No example of your own? Each worksheet has a yellow “Need an example?” box. It gives you ideas to start from, or a made-up company to use instead, so you can still do every step.
- One worksheet at a time: pick one below, or use the numbers in the bar at the top. Each has its own address, so a link in the lab steps opens the right one.
- Keep your work: each worksheet ends with a button that saves it as a PDF with your answers, and a blank fill-in PDF to download. Save your PDFs before you leave today: until then your answers live only in this browser, on this device.
Worksheet 1: Use-case scorecard
Before we start (write this first): How confident am I today that I could sponsor an AI pilot and defend it to finance? (1 = not at all, 5 = fully)
Lab 5, step 3The five criteria
List three candidates, then score each 1–5 on every criterion. Higher is better. Risk is scored in reverse: 5 = low risk.
Lab 5, steps 2–5Candidates Required
Three candidates, each with all five scores. Name each one as a workflow: “triage inbound support tickets,” not “use AI in support.”
First-pilot rule: a first pilot needs 4+ on Risk and 4+ on Measurability, whatever its total.
Quick version, for judging one proposal someone brings you: ask four questions instead of five (fold Data readiness into Feasibility). The rule is the same: Risk and Measurability must pass, however good the Value story is.
Lab 5, step 4Stress-test with AI Optional helper
Before you copy: remove any confidential details. No company, customer or people’s names, and no real figures.
Lab 5, step 6Blocker check Required
A blocker is something that rules a use case out for now, however well it scores. A high score never overrides a blocker. For your top-scored candidate, tick anything that is true. Reading the list is the required part; a tick is only for something that is true:
Lab 5, step 7The winner Required
Press My winner on one of your candidates above.
Why this use case should be your first pilot: one sentence that says what makes it a safe bet.
Lab 5, steps 4–5 and 8Notes from the AI stress-test
The AI reviewer in step 4 asks five hard questions about your top candidate. These boxes hold what you learned from them.
If the AI’s questions made you change one of your five scores: which score, from what to what, and why. Nothing changed? Write “none”.
The one question from the AI reviewer that you could not answer today.
The person who could answer that question: a name or a job title.
Of your winner’s five scores, the one you trust least. You will enter this one in the poll on your phone.
What you would have to check or measure to be sure of that score.
Worksheet 2: Pilot metrics charter
Lab 6, step 2Pilot header
The use case you picked as your winner on Worksheet 1.
The one person who will run the pilot day to day. One person, not a team: a job title, or a name (a made-up one is fine).
The one person who pays for the pilot and can clear roadblocks, often the workflow owner’s manager: a job title, or a name (a made-up one is fine).
The one group of people who will use the AI tool during the pilot, and about how many they are. No names needed.
The first day the pilot team uses the AI tool in real work.
The last day of the pilot: 89 days after the start date (the start date is day 1). On the worksheets page this is worked out for you; change it if your pilot must end on another day.
How you will get the AI tool. Buy: a product that already does this job. Partner: a vendor fits a tool to your data and way of working. Build: your own engineers make it. Most first pilots buy.
Lab 6, steps 3–4Primary metric (pick ONE)
| Family | Example metrics |
|---|---|
| Time | hours per task, cycle time, backlog age |
| Cost | cost per ticket/document/case, contractor spend |
| Quality | error/rework rate, first-pass acceptance |
| Adoption | weekly active users, % of volume through the tool |
The one measurement that should improve if the pilot works. Say what is measured and in what unit (minutes, dollars, percent or a count).
How you will find out what your primary metric is today, before AI: who looks it up, from where, and for which two weeks.
What your primary metric is today, before AI. Not sure? Write your best guess and add “guess”: measuring it is the pilot’s first job.
What you want your primary metric to be by day 90, in the same unit as the baseline value. A number, not “better”.
Lab 6, step 5Quality guardrail
The number that must NOT get worse while the primary metric improves:
A second measurement, one that shows quality. It must not get worse while your primary metric improves.
What your guardrail is today. A guess is fine if you mark it.
Lab 6, step 6Definition of done
One sentence built from the boxes above: the day-90 date, the use case, the primary metric with its baseline value and 90-day target, and the guardrail.
Lab 6, step 7Stop rule
A stop rule is the result that would make you end the pilot early, decided before you start.
The day of the pilot, about halfway through, when you look at the primary metric and the guardrail and decide whether to carry on. A number from 1 to 89.
What the primary metric or the guardrail would have to show on that day for you to stop the pilot early.
Lab 6, steps 8–9Finance red-team findings
A red-team is a deliberate attack on your own plan, to find its weak spots before someone else does. In step 8 an AI reviewer plays a skeptic from finance and names three weak spots, each with a fix. In step 9 you write down the two that matter most.
“Gameable” means the primary metric could look better without the work getting better. Write the weak spot the AI reviewer found in your plan, and how you will close it.
Would the primary metric have improved anyway, without the AI? Write why it might, and how you will tell the difference.
Lab 6, step 10What the pilot needs
For step 10 of Lab 6, after the AI review. What it will take to run the pilot for 90 days: money, people, and a limit on what the AI tool may spend. Rough answers are fine.
What the pilot will cost in money over the 90 days: licenses, usage fees, vendor help. A rough figure is fine.
Who will work on the pilot, and about how many hours a week each.
What the AI tool charges to do the job once, for example one ticket or one document. “Not measured yet” is a fine answer today: it is your first question for the team.
The most you will let the AI tool spend in one month.
The one person who is told when the month’s spending gets close to the cap.
Lab 6, step 10If it succeeds
Also for step 10 of Lab 6. What happens after day 90 if the pilot works: who keeps it going, and what it costs once everyone uses it.
Who keeps the AI tool running once the pilot is over.
What the AI tool will cost when everyone who should use it does, and whose budget covers it.
take-homeSeven questions for the team proposing the pilot Optional
Tick each one you already have a clear answer to. The unticked ones are your next conversation.
Worksheet 3: One-page governance playbook
Lab 7, step 6My call on the PayFast proposal
Required. PayFast is the invoice agent proposed in Lab 7. Decide as if the proposal were on your desk. There is no example here: this one is your call.
Two sentences on why. If you chose “approve with conditions”, name the conditions: what must change before the agent runs.
The seven boxes
Fill in Boxes 1 and 2 in Lab 7. Boxes 3–7 come after them, for back at work. You leave today with a started playbook, and (after Lab 8) a date on every box that is still open.
Lab 7, step 8Box 1 — Inventory (where is AI used today?)
Required: at least two tools. List two or three AI tools already in use where you work. For each one, give the tool and what it is used for, its owner (the one person answerable for it; “nobody yet” is a real answer), and the data it touches (what information it can see). Autonomy tier is how much the tool does on its own. 1 assist: AI drafts, a person does the work. 2 review: AI prepares the action, a named person approves it. 3 autonomous: AI acts, and people check afterwards.
Keep it general. Give owners as job titles, not names. Describe the kind of data (“customer emails”), never the data itself. Describe unapproved use as a habit (“people paste customer emails into personal chatbots”), without saying who does it. No lab step asks you to paste this box into an AI tool, so do not.
Lab 7, step 9Box 2 — Usage policy headlines (three rules you’d publish this week)
A usage policy is the short list of rules for using AI at work. Write three of those rules as headlines: one line each, plain enough to follow without asking anyone.
One rule in one line, for example about which AI tools people may use.
One rule in one line, for example about what information must never go into an AI tool.
One rule in one line, for example about who checks AI work before it is used.
back at workBoxes 3–7: the rest of the playbook (tap to open)
No lab step asks you to fill these in today. In Lab 8 you give each of these five boxes a decide-by date on Worksheet 4: the date by which you will have settled it. Fill in the rest back at work.
back at workBox 3 — Approval path
The approval path is how a new AI idea gets a yes or a no: it is scored on Worksheet 1, someone checks it for risk, and an answer comes back within a set number of days.
Who looks at a new AI idea for risk before it is approved: one person or a small group.
How many days someone with a new AI idea should wait for a yes or a no.
back at workBox 4 — Vendor checks
The questions you will put to every company that sells you an AI tool. Tick the ones you want on your standard list.
back at workBox 5 — Incident path
An incident is AI going wrong: it leaks information, makes something up that gets used, or takes a wrong action. The incident path says who gets the first call.
The one person or desk anyone can call the moment AI leaks, makes things up or misfires.
The date you will walk through a made-up incident to test the path.
back at workBox 6 — Roles
Who looks after AI use: a named owner for each use, plus a small review group that approves new AI ideas and keeps this playbook up to date.
Three to five people from different parts of the organization: job titles are enough.
back at workBox 7 — Audit trail
The audit trail is the record of what your AI tools did, kept so that someone can check afterwards: what gets logged, who reviews it, and how long the logs are kept.
What each AI tool records about its own use: for example who used it, when, and what action it took.
The person or group that looks at those records, and how often.
How long the records are kept before they are deleted.
Worksheet 4: Your 90-day plan
Lab 8, step 1From your earlier worksheets
What you already wrote on Worksheets 2 and 3, so you do not have to look it up or type it again. To change a line, change it on its own worksheet.
Lab 8, steps 2–4Your 30 / 60 / 90
Three blocks of 30 days, counted from today. See what AI is already doing, scope one pilot, then ship it: start the pilot in real work. Boxes marked Required are the ones Lab 8 asks for today; the rest are for back at work.
Lab 8, step 2Days 1–30 — SEE
One task of your own that you will do with AI every working day, and the tool you will use.
The date someone starts listing every AI tool in use: Box 1 of your playbook, done properly.
The one person who will put that list together.
The date your usage policy (Box 2 of your playbook, written out on one page) goes to everyone.
Who will see your scored use cases from Worksheet 1.
Lab 8, step 3Days 31–60 — SCOPE
The pilot charter is Worksheet 2. Write the workflow owner who will sign it off: the same job title or name as on Worksheet 2.
The date the workflow owner signs the charter.
The first day of the two weeks in which you measure your primary metric before AI (the baseline plan on Worksheet 2).
The last day of those two weeks.
The date Boxes 3 and 6 of your playbook are in use.
The date you settle how you will get the AI tool for the pilot.
Lab 8, step 4Days 61–90 — SHIP
The start date on Worksheet 2: the first day the pilot team uses the AI tool in real work.
The day and time each week when you look at the primary metric, the guardrail and how many people are using the tool.
The date of the meeting where you decide to scale the pilot up, change it, or stop it: on or just after the day-90 date on Worksheet 2. Book it now.
Who will hear the result, good or bad, with the numbers.
Day-90 test before you scale — all four must be true. Nothing to tick today: this is the test you apply at that meeting.
Lab 8, step 5A date for each open box of your playbook
Boxes 3–7 of your governance playbook (Worksheet 3) are still open. Give each one a decide-by date: the date by which you will have settled that box back at work.
The date by which you will have agreed how a new AI idea gets a yes or a no.
The date by which your standard list of questions for AI vendors will be agreed.
The date by which you will have agreed who gets the first call when AI goes wrong.
The date by which the review group will be named.
The date by which you will have agreed what gets logged, who reviews it and how long logs are kept.
Lab 8, steps 6–8The announcement paragraph
The short message that tells your manager and team about the pilot. An AI assistant drafts it from your notes in step 6 and tightens it in step 7.
Paste the final paragraph from the AI chat here, after the rewrite in step 7. The lab shows a finished one.
after Lab 8Feedback from the pilot review round
After Lab 8, volunteers describe their pilot in thirty seconds and the room gives one strength and one concern. If you present, write down what you hear.
What people liked about your pilot, in a few words.
The worry someone raised that you will need an answer to.