2026-05-20 · 6 min read
4 AI prompts I use three times a week (as a BI Director)
Four quick ChatGPT and Gemini prompts BI analysts run every week. Then the 6 power-prompts that actually replace busywork, delivered free by email.
I have 247 ChatGPT conversations. A similar number on Gemini. Most are one-offs I opened once and never came back to. The ones I actually use, I open three times a week.
Below are 4 of the lightweight ones, the ones I use as scaffolding while doing real work. They are useful on their own. The 6 heavier prompts (the ones that actually replace 2 hours of busywork each) sit in the free email guide at the bottom.
Prompt 1 of 4
The vague-request triage prompt
When to use it: A PM Slacks "can you pull this?" with no scope, no metric, no time window. You either ask 5 questions and look annoying, or guess and waste an hour.
I am a BI analyst. A stakeholder sent me this request: <paste request>. Their role is <role>. The data I have access to includes <list tables or domains>. Generate the 3 most important clarifying questions I should ask before pulling any numbers. Order them so the most decision-changing question is first. Phrase them politely, not interrogation-style. Do not invent context I did not give you.Why it works: Forces ranking. Most analysts ask 5 questions in a row and look unsure. Three ranked questions makes you look like a senior who has done this before. The "do not invent context" line stops the model hallucinating use-cases the PM never mentioned.
Prompt 2 of 4
The Excel and Google Sheets formula prompt
When to use it: You need a formula for a one-off ad-hoc and you do not want to spend 15 minutes on the INDEX-MATCH-XLOOKUP rabbit hole.
I am in Google Sheets. My data is in columns A to F, headers in row 1. Column A is <description>, B is <description>, etc. I want to <describe goal in plain English>. Give me the exact formula, written for Google Sheets (not Excel). Explain in one sentence what each piece of the formula does. If there is a simpler approach without the formula, tell me that first.Why it works: The "Google Sheets not Excel" specificity prevents the model from giving you XLOOKUP syntax that breaks in Sheets. The "if simpler without formula" line is the unlock. Half the time the answer is a pivot table, not a 9-argument function.
Prompt 3 of 4
The pre-send review prompt
When to use it: You are about to send analysis to leadership and you want one round of adversarial review without bothering a teammate.
I am about to send the analysis below to a Director. Act as a skeptical reviewer. Find the 3 biggest weaknesses in the argument. For each, write the exact follow-up question the Director is most likely to ask. Do not nitpick wording or formatting. Focus on logic gaps, missing controls, alternative explanations I have not addressed. Analysis: <paste>Why it works: The "do not nitpick wording" ban is critical. Without it the model becomes a grammar checker. With it, you get the 3 questions you actually need to pre-answer in your conclusion. I rewrite the conclusion paragraph 90 percent of the time after running this.
Prompt 4 of 4
The estimation prompt
When to use it: Someone asks "how long will this analysis take?" and you have 30 seconds to answer before they assume you are sandbagging.
I have been asked to do this analysis: <paste request>. The data lives in <list sources>. I have <list known constraints, e.g. no access to attribution data>. Break this into the 4 to 6 sub-tasks it actually requires. For each, give me a realistic time estimate in hours assuming I have already started. Add a buffer line at the end for the unknown-unknowns. Output as a markdown table.Why it works: Gets you off the spot. You paste the table back to the stakeholder and the conversation moves from "how long?" to "which sub-tasks can we cut?" That is a much better conversation to be in.
The pattern across all 4
Look at what they have in common:
- Each one is bounded. 3 questions, 4 sub-tasks, specific number, specific format.
- Each one includes real context from your work. The model gets specifics, not abstract instructions.
- Each one ends with an output you can use directly. A table, a formula, three questions. Never a wall of text to read and re-summarise yourself.
- Each one forbids one thing. "Do not invent context." "Do not nitpick wording." Constraints make the output sharper.
If your prompts feel mid, look for these four patterns. Add a number. Add real data. Add an output format. Add one ban. That is the whole skill.
The 6 power-prompts (free email guide)
These 4 prompts are the scaffolding. The 6 heavier ones, the ones that genuinely replace 2 hours of busywork each, are in a free email guide. They cover: SQL debugging walkthroughs, dashboard-to-story summaries, metric drop diagnosis, Monday TL;DR templates, hypothesis-to-experiment design, and the metric definitions that stop arguments forever.
Each one comes with example outputs, the exact paste-into-ChatGPT format, and the one design choice that makes it work instead of returning vague nonsense. One email, no spam.
Free guide
The 6 AI prompts BI analysts actually use
The exact prompt templates BI analysts open 3x a week. Stakeholder rebuttals, SQL debugging, anomaly investigation, slide narratives. Free guide, no card, no spam.
