Stop Guessing. Start Patterning.

Most analysts waste tokens. They ask vague questions, get vague answers, repeat. Fix: **reusable prompt patterns**. Templates that produce consistent, auditable outputs.

Below are battle-tested patterns worth adopting.

---

Pattern 1: **Role Anchoring**

Front-load the model's persona:

```

You are a senior business analyst at a Fortune 500 bank.

Translate the following regulatory change into impacted processes:

```

**Why works:** Constrains vocabulary, depth, and framing. Reduces hallucinated fluff by ~40% in practice.

---

Pattern 2: **Structured Output Pact**

Never accept prose when you need data:

```

Return JSON only:

{

"requirements": [],

"assumptions": [],

"risk_flags": []

}

```

**Why works:** Forces model to categorize. Downstream parsing becomes trivial. Stakeholders get scannable artifacts, not essays.

---

Pattern 3: **Chain-of-Thought Triggering**

Add one line:

```

Think step-by-step before answering.

```

Complex logic, dependency tracing, root-cause analysis — all improve measurably. Especially useful for **requirements decomposition** and **gap analysis**.

---

Pattern 4: **Few-Shot Seeding**

Provide 2–3 examples of desired input/output behavior:

```

Input: "Users report timeout on checkout"

Output: { category: "Performance", severity: "P1", affected_svc: "payment-api" }

Now classify this: "Dashboard fails to load after SSO migration"

```

**Why works:** Eliminates ambiguity. Model mirrors your domain taxonomy instead of inventing one.

---

Pattern 5: **Negative Constraint Framing**

Tell model what NOT to do:

```

Do not propose solutions. Only identify gaps between current and target state.

```

Critical when you need **analysis without prescription** — keeps output actionable for your workflow, not the model's assumptions.

---

Practical Takeaway

| Pattern | Best For |

|---|---|

| Role Anchoring | Stakeholder-ready tone |

| Structured Output Pact | ETL, parsing, dashboards |

| Chain-of-Thought | Complex logic tasks |

| Few-Shot Seeding | Domain-specific classification |

| Negative Constraint | Scope control |

**Next step:** Build a pattern library. One Notion doc, one Markdown file — doesn't matter. Document what works per task type. Reuse relentlessly. Iterate monthly.

Prompt engineering isn't magic. It's **interface design for thinking machines**. Treat it like any other BA skill: practice, refine, share with your team.