Analytics Challenges
Most practice sites test whether you can write SQL. These test whether you can find the real story in messy data and convince someone to act on it — which is the part interviews actually fail people on.
Take a real business problem
Not a puzzle with one clean answer. A messy dataset, a stakeholder with a theory, and a deadline.
Submit your analysis
A public GitHub repo, your answers to the objective questions, and a recommendation short enough for a board to read.
Get scored on both
60% on whether your numbers are right, 40% on whether your analysis would survive a real stakeholder.
Diagnose the E-commerce Revenue Drop
Revenue fell 17% in September at a home-goods retailer. The CEO thinks the market softened. The data says otherwise — find what actually happened and tell them in one page.
Why these are different
Every dataset here is generated with deliberate mess in it — duplicate rows from a retry bug, refunds recorded separately, timestamps in the wrong timezone, inconsistent codes. That is not to be annoying. It is because the gap between a junior analyst and a good one is almost never SQL syntax. It is noticing that the number looks wrong before the stakeholder does.
