What analysts actually spend their time on

It's Wednesday afternoon. The owner of Bisi Provisions, a shop with three branches in Lagos, sends a WhatsApp message with a file attached:

"Please which of our branches is doing best? I need it for Friday's meeting. The sales report is attached."

It sounds like a five-minute job. Add up sales for each branch, and send the answer.

Then you open the file. The first two lines aren't data: they say "Bisi Provisions - Sales Report" and "Printed by: admin". Some dates look like 02/01/2026, others like 2026-01-02, others like 2 Jan 2026. Some amounts say 13500, others "₦13,500". The same branch appears as Yaba, YABA and Yaba with a space after it. Halfway down, a row says TOTAL JANUARY ₦3,115,850. A few rows appear twice.

If you add it up as it is, you'll send the owner a confident, wrong answer. This course teaches you to send the right one, and to explain it so she acts on it.

Words in this lesson

What the job really is

Most job adverts make data analysis sound like building clever charts and models. Real work usually looks more like this:

Notice how much happens before the analysis, in steps 1 to 3, and after it, in steps 5 and 6. People who skip them produce answers that are wrong, or right but ignored.

Why cleaning and checking matter so much

Spreadsheets are powerful, and very easy to get wrong.

In a peer-reviewed paper called Data Organization in Spreadsheets, statisticians Karl Broman and Kara Woo describe the risks. They cite a researcher, Panko, who reported that in 13 audits of real-world spreadsheets, an average of 88% contained errors.

They also describe how Microsoft Excel can turn some text into dates automatically. They cite a study that checked lists of gene names published alongside scientific papers in 18 journals, and found that about 20% of the lists had errors caused by names being converted to dates or numbers.

If scientists publishing in journals get caught out, a sales export from a busy shop certainly can.

What "messy" looks like in the Bisi Provisions file

Throughout this course, you'll practise on a realistic, fictional dataset from Bisi Provisions. The raw sales export for January to March has these problems, and every one would break an analysis:

That TOTAL row deserves special attention, because it's so quiet. Nothing breaks. There's no warning. Your numbers are simply wrong.

Why "₦13,500" isn't a number to a spreadsheet

A spreadsheet stores a cell either as a number it can calculate with, or as text. When the ₦ symbol and commas are typed into the cell itself, the whole thing becomes text.

Microsoft's documentation for Excel says SUM ignores text values and adds only the numeric ones. So a column mixing 13500 and "₦13,500" can give a total that looks sensible, and is missing money. A formula like =A1+B1 that meets text in a cell instead returns a #VALUE! error.

In lesson 2.2 you'll learn how to fix this properly. Formatting a real number to display as ₦13,500 is fine. Typing the ₦ into the cell as text is the problem.

The habit that separates analysts from spreadsheet users

Never edit the raw data.

Broman and Woo put it strongly: your primary data file should contain just the data, with no calculations and no graphs. They recommend keeping it untouched, backing it up, and doing your calculations in a copy.

The reason isn't tidiness. In three weeks, someone will ask, "Where did this number come from?" If you edited the original, you can't show them. An answer you can't trace back to its source is an answer nobody should trust.

You'll set this up properly in lesson 1.2.

What you'll be able to do by the end

• Turn a vague request into a question with a clear answer
• Clean messy exports from POS systems, banks and portals, and record every change
• Answer questions with formulas, pivot tables and SQL
• Think clearly about averages, percentages, growth and samples
• Build a dashboard someone can read without you
• Handle personal data lawfully under Nigeria's data protection law
• Write findings that make busy people act, and a portfolio analysis that gets you hired

Download the practice files

You'll use these throughout the course:

• bisi-provisions-sales-raw.csv: the messy January–March export
• bisi-provisions-sales-clean.csv: six months of clean sales, for later modules
• bisi-provisions-products.csv and bisi-provisions-customers.csv: lookup tables
• bisi-provisions.db: the same data as a database, for the SQL module

All names and figures are made up. Any resemblance to a real business or person is a coincidence.

Common mistakes

The short version

• The job: understand the question → get the data → clean and check → analyse → explain → support a decision.
• Spreadsheets are error-prone: Broman and Woo cite audits where an average of 88% of real-world spreadsheets contained errors.
• Common export problems: title rows, total rows inside the data, numbers stored as text, mixed date formats, inconsistent spellings, duplicates, blank rows.
• A total row inside the data corrupts results silently. In Excel, SUM ignores text, so text amounts quietly go missing.
• Never edit the raw data. Keep it untouched and work in a copy.

Try it yourself

1. Download bisi-provisions-sales-raw.csv.
2. Open it in Google Sheets (File → Import, then upload the file) or Excel. Don't change anything yet.
3. Scroll through the whole file. Write down every problem you notice, and the row number where you first saw it.
4. Find the TOTAL JANUARY row. Write one sentence explaining what would happen to an average sale value if you left it in.
5. In an empty sheet, type 500 in one cell and ₦1,250 in the cell below it. Add them with =SUM( of both cells ), then with = the first cell + the second. Write down what each formula gives you, and why.
6. Find any spreadsheet from your own life, like a bank statement export, a price list or church or club records. List three things in it that would break an analysis.

You'll know you've got it when you can open any export and, before calculating anything, list the problems that would make your answer wrong.