Workshop contents
Session 1 · Prompt Engineering
Prompt Engineering
Asking better questions
A useful prompt usually follows a simple rhythm: prepare, prompt, and polish. First decide what you need from the AI, then write the prompt, then review and improve the result.
Define the objective
Before writing a single prompt, answer four questions:
| Question | What it means |
|---|---|
| Outcome and output | What decision do you need to support, and what form should the deliverable take? |
| Audience | Who will see or act on this output? Their level of detail shapes your prompt. |
| Information to provide | What metrics, documents, or datasets must you feed the AI? It cannot use what it does not have. |
| Scope and limitations | What can AI do here, and where does human expertise need to take over? |
Skipping this planning work is the most common mistake. A well-defined objective is worth more than a clever prompt.
Build the prompt
A prompt is assembled from building blocks. Not every use case needs all of them: the specificity of your desired output determines which to include.
Building blocks
| Block | Purpose |
|---|---|
| Task description | Tell the AI what you want it to do. Be concise and direct. |
| Role | The perspective or expertise the AI should adopt for this task. |
| Boundaries | What to include, what to exclude, and where to stop. |
| Context | The background the AI needs to give a relevant response: business rules, data, constraints, or what has already happened. |
| Specific requirements | Format, length, structure, metrics to include, or standards to follow. |
| Reasoning | Ask the AI to explain its logic, show trade-offs, or justify its recommendation. |
Prompt types
Three levels of prompt structure. Each adds building blocks as the task demands more precision.
| Type | Building blocks | When to use |
|---|---|---|
| Basic | Task description + Context | Quick lookups, simple summaries, first-pass exploration |
| Focused | Task description + Role + Context + Specific requirements | Structured analysis, formatted reports, audience-specific outputs |
| Comprehensive | Task description + Role + Boundaries + Context + Specific requirements + Reasoning | Executive decisions, high-stakes recommendations, cross-functional work |
Prompting tip. Not every use case requires every building block. Start with what your output demands and add blocks only when the result falls short.
Same data, different prompt
You will see the same question asked three ways. The point is not to memorize the prompt. The point is to see how much the answer changes when the instruction becomes clearer.
Example A: Basic prompt
Prompt type: Basic (task description + context)
Using the attached DataCo supply chain dataset, what are the top late delivery products?
Outcome. A loose answer with some product names, but no ranking, no percentages and no clear decision point.
Would you take this to a board meeting?
Example B: Focused prompt
Prompt type: Focused (task description + role + context + specific requirements)
You are a supply chain analyst. Using the attached DataCo supply chain dataset, identify the top 10 product categories with the highest late delivery rate. Output a Markdown table with exactly three columns: Category Name, Total Order Lines, and Late Delivery Rate (%). Sort by late delivery rate descending.
Outcome. A ranked table with real numbers. It looks more useful, but it is still only a sorted list, not yet a decision.
This looks useful. But is it a decision, or just a sorted list?
Example C: Comprehensive prompt
Prompt type: Comprehensive (all six blocks: task description + role + boundaries + context + specific requirements + reasoning)
You are a supply chain analyst presenting to the CEO. Using the DataCo dataset, analyze late delivery patterns and produce: 1. Top 10 product categories by late delivery rate (% of order lines late) 2. For each: the total profit associated with late order lines 3. Compare late-delivery rates across shipping modes 4. One actionable recommendation per category to reduce delays Boundaries: base every finding on the dataset provided. Do not speculate beyond what the data supports. If a pattern is ambiguous, say so. Format as an executive summary with a table. Flag any category where late delivery rate exceeds 50% as CRITICAL. Keep it under 500 words. Before generating the table, think step-by-step in a brief paragraph about how the 50% cutoff compares with the company average. Then apply the cutoff exactly as given. For each recommendation, explain the reasoning: what evidence supports it, what trade-off is involved, and what would need to be true for it to work.
Outcome. A structured executive brief with clearer reasoning. It also exposes two issues that need human judgment before anyone acts.
- 49 of 50 categories are flagged CRITICAL because the company-wide late rate is 54.8% and the cutoff sits below it. The AI followed the instruction precisely. The result is useless because almost everything receives the same warning label.
- Item 3 uncovers the real story. First Class 95% late, Second Class 77%, Standard Class 38%. The faster, more expensive shipping services are the late ones, and this only appeared because the prompt asked for a shipping-mode comparison.
Would you take this to the board now? And what would you still need before acting on it?
Prompt debrief
The gap across the three outputs is visible in seconds: the same AI and the same data can produce a weak paragraph, a useful table, or a decision-style brief depending on how the question is asked.
Key takeawayMost organizations never get past Example A. The same AI and the same data produce radically different results depending on how the question is asked.
Why the 50% cutoff is a trap
The cutoff in Example C is deliberate. It proves a simple point: precise instructions do not fix flawed human logic or missing business context.
| The flawed human assumption | Flag any category above 50% late as CRITICAL. To an executive writing the prompt, 50% sounds like a reasonable bar for a severe problem. |
| What the data actually says | The company-wide late rate is already 54.8%, and 49 of the 50 categories sit between 47.7% and 60.1%. The cutoff was set below the average. |
| The useless output | The AI flags 49 of 50 categories as CRITICAL. When almost everything is critical, the label cannot help anyone prioritize. |
What this proves
- AI can follow a bad rule perfectly.
- A cleaner prompt can improve the format without making the recommendation safe.
- The useful business pattern may still be hidden unless the prompt asks for the right comparison.
Polish and iterate
No first prompt is final. After reviewing the output, refine with follow-up prompts: ask the AI to add what is missing, go deeper on a specific section, or change the format.
Then apply human expertise: check for accuracy, question the assumptions, verify the ethics, and make final edits before the output leaves your hands. AI drafts; you decide.
Prompting is iterative. You can always polish with a follow-up prompt. But the human must know the limitations of the tool, review every output, and own the final result.