The Prompt Engineering Mindset for Data Work
The difference between a useless AI response and a brilliant one is how you ask. Learn the prompting patterns that turn AI into a reliable analytics partner.
The number one skill that separates analysts who get value from AI and those who don't isn't technical — it's communication. The way you phrase your request determines whether you get a useless generic response or exactly what you need.
The Golden Rule: Context + Task + Format
Every great prompt has three parts: who you are and what you're working with (context), what you need done (task), and how you want the answer (format). Miss any one of these and you'll get a mediocre response.
Manual Workflow
With AI
Pattern 1: Role Setting
Tell the AI who you are and who it should be. This dramatically improves the relevance of responses.
Set the role before asking your question:
You are a senior marketing analytics consultant. I'm a mid-level marketing analyst at a B2B SaaS company. Our main KPIs are MQLs, pipeline generated, and CAC payback period. I need to build a weekly executive dashboard. What metrics should I include, and how should I structure it?
Pattern 2: Show, Don't Tell
Instead of describing your data, show a sample. AI understands examples much better than descriptions.
Include a sample of your actual data:
Here's a sample of my campaign data: Date, Channel, Spend, Clicks, Conversions, Revenue 2024-03-01, Google Ads, 1250.00, 342, 28, 4200.00 2024-03-01, Meta Ads, 890.00, 567, 15, 2100.00 2024-03-01, LinkedIn, 2100.00, 89, 8, 6400.00 Write a SQL query that identifies which channels are underperforming relative to spend. Define 'underperforming' as ROAS below 2.0.
Pattern 3: Chain of Thought
For complex analysis, ask the AI to think step-by-step. This produces more accurate and thorough results.
Ask for step-by-step reasoning:
I need to analyze why our email channel conversion rate dropped 40% last month. Think through this step by step: 1. What data would I need to investigate this? 2. What are the most likely causes? 3. What SQL queries would help me diagnose each cause? 4. How should I present findings to my VP of Marketing?
Pattern 4: Specify the Output Format
Don't let the AI decide how to format the response. Tell it exactly what you need.
Be explicit about the format you want:
Analyze this campaign performance data and give me: 1. A one-paragraph executive summary (no jargon, suitable for a CEO) 2. A table showing top 5 and bottom 5 campaigns by ROAS 3. Three specific, actionable recommendations with expected impact 4. The SQL query I can run to monitor this weekly Keep the total response under 500 words.
Take a vague question you'd normally ask AI and rewrite it using the Context + Task + Format pattern:
Rewrite this vague prompt into a specific one using the Context + Task + Format pattern: Vague: "Help me analyze my marketing data" Rewrite it by adding: - Context: Your role, company type, data structure - Task: Specific analysis you need - Format: How you want the output structured Then try both versions and compare the responses.
Next up: the iteration loop — the workflow pattern that makes AI a reliable partner instead of a frustrating guessing game.
Get weekly job alerts
Curated marketing analytics roles — delivered every Monday.