Pick a depth. Each prompt opens in your AI pre-loaded with the lesson. Click a row to preview the prompt.
Toy datasets make toy skills. If you practice on a pretend example, you never hit the real snags - a column with mixed text and numbers, a date stored as text, a total that does not match reality - and those snags are 80 percent of the actual job. Choosing a real dataset now, one you care about, means every exercise in this course produces something you could actually use or show someone. It also forces the first real judgment call of analytics: is this data even shaped so a spreadsheet can work with it? Getting that right at the start saves you from discovering, three tasks in, that your source is a PDF of scanned receipts that no tool can slice.
You have three good sources, in order of preference. First, your own or your employer's data - a sales export from Shopify or Square, a CSV of orders, a download from your bank, a Google Form's responses. This is best because you already know what the answer should roughly be, which makes sanity-checking easy. Second, a public dataset on a topic you care about. Our World in Data (ourworldindata.org) lets you download clean CSVs on almost anything - CO2 emissions by country, life expectancy, energy use - via a 'Download' button under each chart. Kaggle (kaggle.com/datasets) has thousands of business datasets - Netflix titles, retail sales, Airbnb listings - free with a login. Government open-data portals (data.gov in the US, data.gov.uk in the UK) publish everything from crime stats to restaurant inspections. Now the crucial judgment: a workable dataset is TABULAR, meaning every ROW is one record (one sale, one country, one movie) and every COLUMN is one field (date, price, category). Here is a workable example - a sales CSV where each row is one order: columns Order Date, Product, Category, Quantity, Revenue, Customer Region. Here is an UNworkable one - a PDF report with paragraphs of text and a summary table buried in an image; you cannot slice that. If your source is a scanned document, a website you would have to copy by hand, or one giant merged cell, pick something else now. Aim for something with at least 50 rows and 3 or more columns so there is genuinely something to slice.