Capstone: analyzing a real dataset end to end
Time to put it all together. Before you is one dataset — 2400 users with revenue, device, A/B group and whether they purchased. We'll walk it through the whole analyst's path: from the first glance to an honest decision. The widget is step-by-step: click through stages 1–6 and watch each tool of the course come into play.
2400 users, revenue per user. The distribution is right-skewed: the mean (19) is noticeably above the median (11). Takeaway: the typical user is more honestly described by the median than by the mean.
Steps 1–2 in the widget — getting to know the data. The first thing an analyst does: look at the distribution, not compute the mean blindly. Revenue is skewed, the mean is above the median → we describe it with the median. The outliers are real (large purchases) → we don't remove them, but use the mean cautiously. This is the "Descriptive" and "Distributions" modules in action.
The analyst's main skill is not knowing the formulas but leading the data through the whole cycle: look at the shape, clean it, slice by segment, choose an appropriate method, check the assumptions and state an honest conclusion. Any single method is useless in isolation from this chain.
Take your own data (expenses, game stats, a work dataset) and walk the same route: distribution → outliers → segments → relationship/comparison → conclusion. Understanding sticks when the sliders and steps are on your own questions to real numbers.
This is how the daily work of a product and data analyst is built: 80% of the time is inspecting, cleaning and exploring the data, and only then the method. Skipping EDA is the most common cause of wrong conclusions.
And this is how strong write-ups are built too: not "I know the t-test", but "here is how I led the data from question to decision, what I checked and where I honestly flagged the uncertainty".