Choose an executive decision

Outcome and preparation

Day 1 · 09:00–09:30 · 30 minutes. Choose one recurring decision and establish a baseline you can compare on Day 2. You need a signed-in Claude App and the practice pack. Complete PREWORK.md first. Spend 5 minutes on the example, 20 practising and 5 checking.

Northstar's CEO is deciding whether to fund reliability and support work before increasing acquisition spend. Its CTO needs evidence about errors, latency and incidents. Both need the same facts, presented for different decisions. A report is useful when it changes what someone can decide—not merely when it sounds polished.

Step by step

1. Step 1

   Open a new Claude conversation. Name it Northstar — decision brief. Open reference/strategy.md and reference/system-overview.md locally; confirm both say fictional. Do not upload the facilitator answer key.

2. Step 2

   Write your task card using the prompt below. For this exercise use the Northstar decision; add your own recurring task separately if its inputs are approved.

   Prompt / exampleCopy

   Help me define a recurring executive report, without drafting the report yet.
   Organisation: Northstar Services, a fictional support software company.
   Audience: CEO and CTO.
   Decision: should we fund a four-week reliability and support pilot before expanding acquisition?
   Inputs available: two weeks of business KPIs, service health, incidents and strategy.
   Ask me one question at a time about cadence, required evidence, constraints and reviewer.
   Finish with: decision, audience, input files, period, output, owner, reviewer and success criteria.
   Do not invent business facts or imply that the pilot budget has been approved.
   

3. Step 3

   Answer the questions. Set cadence to weekly, output to a one-page draft plus calculation appendix, owner to yourself and reviewer to a named colleague or facilitator. Save the resulting text as handover/task-card.md in your own handover folder.

4. Step 4

   Record your baseline: preparation minutes, checking minutes, runs per month and current error/rework observations. If you have no measured baseline, write estimate — to measure, not a precise claim. Start a timer for today's exercise and save it in handover/rehearsal-log.md.

5. Step 5

   Define success in observable terms: correct period, correct calculations, a source for each factual claim, visible unknowns, clear options and human approval before distribution. Ask Claude to critique the task card.

   Prompt / exampleCopy

   Review this task card. Identify anything a second analyst could interpret differently.
   Suggest one correction per ambiguity. Do not change the reporting scope or add automation yet.
   [Paste task card]
   

6. Step 6

   Revise the card and explain it to a partner in two minutes. Ask them what decision the report supports and which missing source would stop it. Save their feedback and your revision.

Expected output and verification

Your task card identifies one decision, a weekly cadence, approved sources, a report shape and a reviewer. It does not say only “use AI to improve productivity.” A partner should be able to describe what is in and out of scope without asking you to explain it again.

Check that a preparation-time estimate is labelled as an estimate. The later ROI exercise needs both preparation and review time; excluding review would overstate savings.

Troubleshooting

• The task is too broad: choose one decision and one report. “Run the company” is not an MVP.
• You cannot use company data: finish with Northstar. Real data is optional throughout.
• Claude starts writing the report: remind it that the current output is the task card only.
• No baseline exists: time one manual attempt or keep the uncertainty explicit.

Completion checklist

• task-card.md saved with decision, sources, owner and reviewer.
• Baseline recorded with measured/estimated labels.
• Partner can explain the success criteria.

Next: choose the simplest AI approach that can meet this task card.