Prompt Exercise — Executive Decision Brief

Why This Exercise Matters for Leaders

You don't write risk assessments — you make decisions based on them. This exercise teaches you to write prompts that produce the kind of output you actually need: clear recommendations, quantified risks, and actionable next steps.

	What Leaders Do	Prompt Technique You'll Learn
📋	Write executive summaries for board	Structured output + persona
⚖️	Make approve/reject/escalate decisions	Chain-of-Thought + grounding
✅	Review team output for quality	LLM-as-Judge rubric
👥	Communicate to different audiences (board vs ops vs regulators)	Same data, different personas
🎯	Decide with incomplete information	Self-consistency + guardrails

The Scenario

You're a VP of Financial Risk at AnyCompany. Your risk analytics team has flagged a merchant — QuickMart Express — with an AMBER risk rating. The merchant's PayLater credit line is SGD 50,000. You need to decide: Approve, Restrict, Suspend, or Investigate?

You'll write a prompt that helps you produce a 1-page Executive Decision Brief for the CFO — the kind of document that gets a decision made in one meeting.

🚫 No copy-paste prompts

Each step gives you hints and requirements — you write the actual prompt yourself in Claude. This forces you to think about what makes a good prompt. If you're stuck, unlock the reference prompt with the instructor password.

⚙️ How it works

• Start a new conversation in Claude for each step (Steps 1–5)
• Step 6 continues in the same conversation as Step 5
• The merchant data is provided — copy it once, paste into each prompt
• At the end, submit your best output for AI Judge scoring — leaderboard shows who's winning

Techniques You'll Practice

Step	Technique	What It Adds	Time
1	Zero-Shot Baseline	See what "no guidance" produces	5 min
2	Role & Persona	Executive voice and decision framing	8 min
3	Structured Output	Scannable brief format	8 min
4	Chain-of-Thought + Grounding	Show reasoning, cite data	10 min
5	Few-Shot + Self-Critique	Consistent quality, catches errors	10 min
6	Template Extraction	Reusable for any decision	10 min
—	AI Judge Competition	Score and compete	15 min

1Zero-Shot Baseline

Step 1: Zero-Shot — Just Ask

📖 Technique: Zero-Shot Prompting
Give the model a task with no examples, no role, and minimal instruction. This establishes your baseline — you'll see what "no guidance" produces, then improve from there.

💡 Your Task — Write a prompt that:

Asks the AI to help you decide what to do about this merchant
Provides the merchant risk assessment data (copy button below)
Uses no more than 1–2 sentences of instruction
Does NOT specify a role, format, or structure

Think: What's the simplest way to ask "what should I do about this merchant?"

Merchant Risk Assessment — QuickMart Express

🔍 After you get the output, observe:

Would you send this to your CFO? Why not?
Is there a clear recommendation (A/B/C/D)?
Does it quantify the financial impact?
Could someone act on this without asking follow-up questions?

🔑 Reveal reference prompt (instructor password)

Reference Prompt — Step 1

Based on this merchant risk assessment, what should we do? Should we approve, restrict, suspend, or investigate? [PASTE MERCHANT DATA]

💡 Notice how minimal this is. The output will be generic, unstructured, and probably too long. That's the point — this is your "before" baseline.

2Role & Persona

Step 2: Add Your Executive Persona

📖 Technique: Role & Persona Prompting
Assigning a role changes what the AI considers important. A "VP of Risk" focuses on financial exposure and governance. A "customer support lead" would focus on complaint resolution. The persona shapes the entire output.

💡 Your Task — Write a prompt that includes:

A persona — who are you? (VP of Financial Risk? Head of Merchant Operations?)
Give them experience — how many years? What region?
Give them a decision style — are they conservative? data-driven? balanced?
Specify the audience — who will read this? (CFO? Risk Committee?)
Then ask for the decision recommendation with the merchant data

Think: When you write a decision memo, what voice do you use? What does your CFO expect to see?

Merchant Data

🔍 Compare with Step 1: The output should now sound like an executive — concise, decision-focused, with financial framing. It should feel like something you'd actually send upward.

🔑 Reveal reference prompt

Reference Prompt — Step 2

You are a VP of Financial Risk at a Southeast Asian fintech company. You have 10 years of experience managing merchant credit risk across 6 markets (Singapore, Malaysia, Indonesia, Thailand, Vietnam, Philippines). You are known for being data-driven and balanced — you protect the company from losses while supporting legitimate merchant growth. Your audience is the CFO, who has 5 minutes to read your recommendation and make a decision. Based on the merchant risk assessment below, write a decision recommendation: should we Approve (maintain current terms), Restrict (reduce credit line), Suspend (pause PayLater), or Investigate (formal fraud review)? [PASTE MERCHANT DATA]

💡 The persona + audience framing changes everything. "VP writing for CFO" produces concise, decision-ready output. "Analyst writing a report" produces long, detailed analysis.

3Structured Output

Step 3: Define the Brief Format

📖 Technique: Structured Output
Defining exact sections ensures every decision brief covers the same areas. Your CFO can scan it in 2 minutes because they know exactly where to find the recommendation, the risk, and the financial impact.

💡 Your Task — Specify EXACTLY what sections the brief should have:

Think about what a CFO needs to make a decision in one meeting. Consider:

Decision — your recommendation (A/B/C/D) in one sentence, upfront
Situation Summary — what's happening, in 3 sentences max
Financial Exposure — how much money is at risk, under each option
Key Evidence — the 3-4 data points that drive your decision
What Could Change My Mind — what additional info would flip the decision
Conditions & Timeline — what happens next, review date, escalation triggers

Think: If your CFO only reads the first 3 lines, do they know the decision? If they read the full page, can they defend it to the board?

Merchant Data

🔍 Compare with Step 2: The output should now be scannable — clear sections, decision upfront, financial numbers visible without reading the whole thing. A CFO could act on this in 2 minutes.

🔑 Reveal reference prompt

Reference Prompt — Step 3

You are a VP of Financial Risk at a Southeast Asian fintech. Your audience is the CFO (5 minutes to read and decide). Produce a 1-page EXECUTIVE DECISION BRIEF with EXACTLY these sections: 1. RECOMMENDATION (one sentence — Approve/Restrict/Suspend/Investigate + one-line justification) 2. SITUATION (3 sentences max — what's happening with this merchant) 3. FINANCIAL EXPOSURE - Current exposure (SGD) - Monthly loss at current trajectory - Exposure under each option (A/B/C/D) 4. KEY EVIDENCE (bullet list — the 3-4 data points that drive your recommendation) 5. MITIGATING FACTORS (bullet list — legitimate explanations that reduce concern) 6. WHAT WOULD CHANGE MY MIND (what new information would flip the decision) 7. CONDITIONS & NEXT STEPS - Immediate actions (this week) - Review date - Escalation trigger (what happens → auto-escalate to next level) Keep the entire brief under 400 words. Decision-makers scan, they don't read essays. [PASTE MERCHANT DATA]

4Chain-of-Thought + Grounding

Step 4: Show Your Reasoning & Cite the Data

📖 Technique: Chain-of-Thought + Grounding Rules
Chain-of-Thought makes the AI show its reasoning before concluding — like showing your work in math. Grounding rules prevent hallucination by requiring every claim to cite specific data. Together, they produce output that's auditable and defensible.

💡 Your Task — Add these two elements:

Chain-of-Thought: Tell the AI to reason through the decision before stating it:

First, list the factors that support a more severe action (Suspend/Investigate)
Then, list the factors that support a less severe action (Approve/Restrict)
Then, weigh them against each other
THEN state the recommendation

Grounding Rules: Tell the AI:

Every claim must reference a specific number from the data
Do NOT assume facts not in the assessment
If information is missing, say so — don't guess
Distinguish between "concerning pattern" and "confirmed fraud"

Think: If the board asks "why did you recommend this?" — can you point to specific data for every claim?

Merchant Data

🔍 Compare with Step 3: The reasoning is now visible and auditable. Every claim cites a number. The CFO can see WHY you recommend this option — not just what you recommend. This is governance in action.

💬 Why this matters for governance: The customer's #1 request was "how do we verify and trust AI outputs?" Chain-of-Thought + Grounding is the answer. When the AI shows its reasoning AND cites specific data, you can verify each step. If the reasoning is wrong, you catch it. If the data citation is incorrect, you catch it. This is how you build trust in AI-assisted decisions.

🔑 Reveal reference prompt

Reference Prompt — Step 4

You are a VP of Financial Risk at a Southeast Asian fintech. Audience: CFO (5 min to read). GROUNDING RULES: - Every factual claim must cite a specific number from the data below - Do NOT assume information not present in the assessment - If data is insufficient to assess something, state: "[NEED: specific data required]" - Distinguish between "concerning pattern" and "confirmed fraud" — do not conflate them REASONING APPROACH: Before stating your recommendation, think through: 1. List factors supporting MORE severe action (Suspend/Investigate) — with data citations 2. List factors supporting LESS severe action (Approve/Restrict) — with data citations 3. Weigh: which set is stronger? What's the cost of being wrong in each direction? 4. THEN state your recommendation Produce a 1-page EXECUTIVE DECISION BRIEF with these sections: 1. RECOMMENDATION (one sentence + justification) 2. REASONING (the weighing above — keep to 4-5 bullet points) 3. FINANCIAL EXPOSURE (current + projected under each option) 4. KEY EVIDENCE (3-4 data points, each with exact numbers) 5. MITIGATING FACTORS (legitimate explanations) 6. WHAT WOULD CHANGE MY MIND 7. CONDITIONS & NEXT STEPS (actions, review date, escalation trigger) Under 500 words total. [PASTE MERCHANT DATA]

5Few-Shot + Self-Critique

Step 5: Add Examples & Self-Review

📖 Technique: Few-Shot Examples + Self-Critique
Examples teach the AI your exact tone and depth. Self-critique makes it review its own output before presenting — like having a colleague proofread your memo before it goes to the CFO.

💡 Your Task — Add these two elements:

Few-Shot Example: Write a short (3-4 sentence) example of what a GOOD decision brief recommendation looks like. Show the tone, the specificity, the format you want.

Include one example of a "Restrict" recommendation (since that's the likely answer here)
Show how it cites numbers, states conditions, and sets a review date

Self-Critique: After generating the brief, tell the AI to:

Re-read the brief as if you're the CFO receiving it
Check: Is the recommendation clear in the first 10 seconds?
Check: Are all financial numbers accurate and sourced from the data?
Check: Would you approve this without asking follow-up questions?
Fix any issues before presenting the final version

Merchant Data

🔍 This is your best output. The combination of persona + structure + CoT + grounding + example + self-critique should produce a brief you'd actually send to your CFO. If it's not there yet, iterate — adjust the prompt and try again.

🔑 Reveal reference prompt

Reference Prompt — Step 5 (full production prompt)

You are a VP of Financial Risk at a Southeast Asian fintech. Audience: CFO (5 min to read and decide). EXAMPLE of the tone and depth I want: --- RECOMMENDATION: RESTRICT — Reduce PayLater limit from SGD 80K to SGD 40K pending 60-day review. REASONING: Chargeback rate of 3.2% (3x benchmark) indicates operational strain, but 2 new outlets opened in the same period and merchant has 18 months of clean history. Pattern is more consistent with growing pains than fraud. Restricting (not suspending) protects exposure while preserving the merchant relationship. FINANCIAL EXPOSURE: Current SGD 62K outstanding. At current chargeback rate, projected monthly loss SGD 4.1K. Under restriction, exposure caps at SGD 40K (loss capped at SGD 2.6K/month). ESCALATION: If chargeback rate exceeds 5% in next 30 days → auto-escalate to Suspend. --- GROUNDING RULES: - Every claim must cite a specific number from the data - Do NOT assume facts not in the assessment - Distinguish "concerning pattern" from "confirmed fraud" - If data is missing, state what you need REASONING: Before recommending, weigh factors for/against severity. Show your work. SELF-REVIEW: After drafting, re-read as the CFO: - Is the decision clear in 10 seconds? - Are all numbers accurate and from the data? - Would you approve without follow-up questions? - Fix any issues before presenting. Produce a 1-page EXECUTIVE DECISION BRIEF: 1. RECOMMENDATION (one sentence + justification) 2. SITUATION (3 sentences max) 3. REASONING (for vs against, then weighing) 4. FINANCIAL EXPOSURE (current + each option) 5. KEY EVIDENCE (3-4 bullets with numbers) 6. WHAT WOULD CHANGE MY MIND 7. CONDITIONS & NEXT STEPS (actions, review date, escalation) Under 500 words. [PASTE MERCHANT DATA]

6Template Extraction

Step 6: Make It Reusable

📖 Technique: Meta-Prompting (Template Extraction)
Ask the AI to turn your refined prompt into a reusable template with {{variables}}. Any decision — not just this merchant — can use the same template. This is how one 90-minute exercise saves your team hundreds of hours.

In the same conversation as Step 5 (don't start a new one), type this follow-up:

💡 Your Task — Ask the AI to create a reusable template that:

Works for ANY merchant decision — not just QuickMart Express
Uses {{variables}} for all merchant-specific data
Includes the persona, grounding rules, reasoning approach, structure, and self-review
Has a brief usage guide: "paste this template, fill in the data, get a decision brief"

Think: If you handed this to a colleague who missed today's workshop, could they use it immediately?

✅ Deliverable: You now have a reusable Executive Decision Brief template. This is your Day 1 artifact — copy it somewhere safe. Tomorrow on Day 2, you'll convert this into a Claude Cowork project (steering + skill).

🔑 Reveal reference prompt

Reference Prompt — Step 6

Now turn this into a reusable template that any VP or director on my team can use for ANY merchant credit decision — not just QuickMart Express. Create a complete prompt template that: 1. Uses {{variable_name}} placeholders for all merchant-specific data 2. Includes the persona, grounding rules, reasoning approach, brief structure, and self-review 3. Has a USAGE GUIDE at the top: - What data to gather before using (checklist) - How to fill in the variables - Expected output format and length 4. Works for all 4 decision types (Approve/Restrict/Suspend/Investigate) The template must be self-contained — a colleague who wasn't in this workshop should be able to use it immediately.

🏆 AI Judge Competition — Score Your Brief

Submit your best Executive Decision Brief (from Step 5) for AI scoring. The AI Judge evaluates on 4 dimensions that matter for executive communication.

Evaluation Rubric

Criteria	Score 1–2	Score 3–4	Score 5
Decision Clarity	Recommendation buried or ambiguous	Clear recommendation with some justification	Decision obvious in first 10 seconds, fully justified
Risk Quantification	Vague ("significant risk")	Some numbers cited	Every claim has a specific SGD amount or percentage from the data
Reasoning Quality	Just states opinion	Shows some weighing of factors	Clear for/against analysis, distinguishes pattern from proof, acknowledges uncertainty
Actionability	Generic ("monitor closely")	Some next steps listed	Specific actions + owners + timeline + escalation trigger

🚀 Submit Your Decision Brief for Scoring

Your Name Paste your Executive Decision Brief (from Step 5)

DIY Scoring — Paste into Claude

Alternatively, copy this AI Judge prompt and paste it into a new Claude conversation with your brief:

AI Judge Prompt — Copy into Claude

You are a STRICT evaluator of executive decision briefs for a fintech CFO. You have high standards — a score of 5 should be genuinely exceptional. Most good briefs score 14-17/20. Score the following Executive Decision Brief on 4 criteria (1-5 each): 1. **Decision Clarity** (1-5): Is the recommendation immediately obvious? Could the CFO act in 10 seconds? 2. **Risk Quantification** (1-5): Are financial impacts stated in specific SGD amounts with data citations? 3. **Reasoning Quality** (1-5): Does it show clear for/against weighing? Distinguish pattern from proof? 4. **Actionability** (1-5): Are next steps specific with timeline, owner, and escalation trigger? Return as JSON: {"clarity": X, "quantification": X, "reasoning": X, "actionability": X, "total": X, "strengths": "one sentence", "weakness": "one sentence — there is ALWAYS something to improve"} DECISION BRIEF TO EVALUATE: [PASTE YOUR BRIEF HERE]

🎯 What you built today:

Through 6 iterative steps, you evolved a 10-word question into a production-grade decision template. The techniques you practiced:

1. Zero-shot → showed what "bad" looks like
2. Persona → executive voice and decision framing
3. Structure → scannable, consistent format
4. CoT + Grounding → auditable reasoning, no hallucination
5. Few-shot + Self-critique → consistent quality, catches errors
6. Template extraction → reusable for any decision

Tomorrow (Day 2): You'll convert this template into a Claude Cowork project with steering rules — turning it from a prompt you paste into an agent that runs automatically.

📋 Executive Decision Brief

Why This Exercise Matters for Leaders

The Scenario

Techniques You'll Practice

Step 1: Zero-Shot — Just Ask

💡 Your Task — Write a prompt that:

Step 2: Add Your Executive Persona

💡 Your Task — Write a prompt that includes:

Step 3: Define the Brief Format

💡 Your Task — Specify EXACTLY what sections the brief should have:

Step 4: Show Your Reasoning & Cite the Data

💡 Your Task — Add these two elements:

Step 5: Add Examples & Self-Review

💡 Your Task — Add these two elements:

Step 6: Make It Reusable

💡 Your Task — Ask the AI to create a reusable template that:

🏆 AI Judge Competition — Score Your Brief

Evaluation Rubric

🚀 Submit Your Decision Brief for Scoring

DIY Scoring — Paste into Claude