How to Analyze Data With AI (When You Are Not a Data Person)

There is a specific kind of dread that comes from a spreadsheet someone forwarded you with the note "any thoughts?". You can read the numbers fine. What you cannot do quickly is say what they mean, and that is the part your boss actually wants.

AI closed that gap faster than almost any other office skill. You upload the file, ask the right questions, and get back the story in the data plus the chart to show it. This guide covers the questions worth asking, how to make charts without formula knowledge, and the checks that keep you from confidently repeating something wrong.

A note on scope: this is about understanding data. If what you need is a formula or a cleanup step inside the sheet itself, our plain-English Excel formulas guide is the right place.

Step 1: upload, then ask for orientation

Every major assistant can read a spreadsheet or CSV directly. Attach the file and start with orientation rather than conclusions, because a tool that has not described your data will happily analyze the wrong column:

Here is my data file. Before analyzing anything, describe it back to me: what each column contains, how many rows, the date range if any, and anything that looks messy (blanks, duplicates, inconsistent formats, outliers). Then tell me 5 questions this data could actually answer, and 3 it cannot.

That last pair is the sleeper feature. Knowing what your data cannot answer prevents the most common analysis mistake, which is confidently answering a question your numbers do not support.

Step 2: ask business questions, not data questions

"What is the average?" gets you a number nobody asked for. Ask the question your manager is really asking:

Answer these in plain language, showing which numbers support each answer: 1) What is the single most important thing this data says? 2) What changed compared to the earlier period, and by how much? 3) Which [customers / products / regions / weeks] are performing unusually well or badly? 4) What is surprising here that I would probably miss? 5) What should I look into next? No jargon. Explain any statistical term you use in one sentence.

Then follow the thread. Analysis is a conversation: "why is March low?", "is that a trend or one bad week?", "exclude the outlier and redo it". The follow-ups are where the real insight lives, exactly as with summarizing long documents.

Step 3: get a chart you can actually put in a deck

Assistants with data analysis can generate charts from your file directly, and they are better at picking the right chart type than most of us:

Make one chart that best shows [the finding]. Pick the chart type that communicates it most clearly and tell me why you chose it over the alternatives. Keep it simple: no 3D, no dual axes, no more than [5] categories. Then give me a one-sentence caption stating the takeaway, not describing the chart.

The caption instruction matters. "Revenue by month" is a label; "Revenue grew every month except March, when the outage hit" is a point. Charts with points get remembered, which also makes them ideal for the presentation you will build next.

Step 4: check before you repeat it

Data analysis has a specific failure mode: the AI misreads a column, silently drops rows with blanks, or treats text-formatted numbers as zero, and the result looks perfectly plausible. So verify the spine of the analysis, not every cell:

Show me exactly how you calculated each number in your summary: which columns and rows you used, how you handled blanks and duplicates, and any rows you excluded. Then flag anything in this data that could make your conclusions wrong.

Then spot-check one figure by hand. If the total matches your own count and one segment matches your gut, you are almost certainly safe. If a number surprises you, chase it before you present it, because confident wrongness is the risk here.

One rule about the file you upload

Real business data often contains customer names, emails, salaries, or unreleased figures. Delete or anonymize those columns before uploading, or use your company's approved AI workspace. Analysis rarely needs identities: replace names with Customer A, B, C and the insight is identical. The full rules live in our workplace safety guide.

You do not need to become an analyst

You need to walk into the meeting able to say what the numbers mean, what changed, and what you would do about it. That was always the job. AI just removed the two hours of wrestling with pivot tables that used to sit in front of it.

Working through your own real numbers is the fastest way to learn this. Our free course Everyday Productivity: Get Hours Back Every Week builds the habit alongside planning, research, and summarizing.