From Messy Data to Clear Recommendation in Under 5 Minutes

You have a spreadsheet. It has 47 columns, 2,000 rows, and no obvious story. Your boss wants “the key takeaway” by end of day. You know the answer is in the data somewhere. You just cannot see it yet.

So you do what everyone does: you sort by one column, then another. You highlight a few cells. You make a pivot table that does not quite answer the question. You make another one. Two hours later you have 6 tabs open and you are less sure of the answer than when you started.

This is Issue #6 of The AI Playbook. Today I am going to show you how to extract a clear recommendation from messy data in under 5 minutes — using prompts that do the analysis your pivot tables cannot.

Why Spreadsheets Are Hard (and AI Fixes It)

The problem with data analysis is not the math. It is the framing. Most people open a spreadsheet and ask “what does this say?” — which is the wrong question. The right question is: “What decision does this data help me make?”

Spreadsheets show you everything. AI can focus on what matters.

The Pattern Finder (90 seconds)

Paste your data — CSV, table, or even a screenshot — and use this prompt:

Here is a dataset with [X rows] and [X columns]:

[paste data or describe columns]

I need to make a decision about [what you are deciding].

Analyze this data and tell me:

1. The 3 most important patterns or trends

2. Any outliers or anomalies that would change the conclusion

3. What this data suggests I should do — in one sentence

Real example: I pasted 6 months of customer churn data — 1,800 rows, 12 columns. Within 90 seconds, the AI identified that churn was concentrated in accounts onboarded by one specific team, during one specific quarter. Nobody had caught this because the spreadsheet showed the average churn rate (low), not the distribution (bimodal). The recommendation was obvious once the pattern was visible.

The Assumption Checker (60 seconds)

This is the prompt most people skip. It is also the most important one. AI is confident by default — this forces it to show you where the uncertainty is.

Based on your analysis, what assumptions are you making?

For each assumption:

- How confident are you (high/medium/low)?

- What data would I need to verify it?

- If this assumption is wrong, how would it change your recommendation?

The key insight: AI will give you a clean recommendation even when the data is ambiguous. This prompt forces it to show you where the uncertainty is — which is exactly where the real risk hides.

The Visualization Plan (90 seconds)

Most people think visualization is the final step. It is actually the communication step. You already have the answer — this prompt helps you show it to someone else in a way they grasp in 5 seconds.

Create a simple visualization plan for this data. Tell me:

1. What chart type would best show the key pattern you found?

2. What should be on the X axis, Y axis, and any grouping?

3. Write the formula or code (Excel/Google Sheets/Python) to create it.

I want ONE chart that I could put on a slide and have my audience immediately understand the conclusion.

Why it works: One chart that tells the whole story is worth more than a 20-slide deck. This prompt gives you the chart type, the axes, and the code to build it.

The Full System

1. Paste data + decision context — what are you deciding? (30 sec)
2. Run Prompt 1 — get patterns, outliers, and recommendation (90 sec)
3. Run Prompt 2 — check assumptions, find uncertainty (60 sec)
4. Run Prompt 3 — get the one chart that tells the story (90 sec)
5. Add your judgment — does the recommendation match your domain knowledge? (60 sec)

Total time: under 5 minutes. Compare that to 2 hours of pivot table archaeology.

What This Works For

The prompts are the same. The data changes. The speed does not.

The Trap to Avoid

Do not use AI analysis as the final word. Use it as the first draft. AI is excellent at finding patterns and terrible at understanding context. It does not know that your CFO hates pie charts, that Q3 was an anomaly because of a one-time contract, or that the “outlier” account is actually your biggest customer who just had an unusual month.

Your job is the last mile: take the AI’s pattern recognition and apply your domain knowledge. That combination — machine pattern-finding plus human judgment — is where the real leverage is.

2 hours of pivot tables Replaced with 5 minutes of prompts

Decision-ready recommendation — not a data summary.

Try It Today

Pick a spreadsheet you have been staring at. Any one. Run the 3 prompts. Pay attention to how fast you go from “I have data” to “I have a recommendation.”

Then reply to this email and tell me how it went. I read every response.

Persistent AI Context: How to build an AI research assistant that actually remembers what you told it last week — the prompt architecture for persistent AI memory.