← All work
~300
Entrants in the challenge
3 + 1
Top products, overall and per city
4
Steps from raw file to answer

A LinkedIn data-analysis tournament: find the top 3 products overall, and the top product in each city.

Project Type
Competition
Stack
PythonPower BIExcel
Year
2025

Case study

  1. Overview

    Sales Analytics Challenge, an open LinkedIn competition with around 300 entrants. I entered on my own as the analyst, cleaning in Python and building the dashboard in Power BI from a single raw Excel file.

  2. Problem

    The brief asked for the top 3 products by sales and the leading product in each city, plus visuals that support a decision. The supplied file could not answer it as it stood, because missing values and duplicates sat under every number, so any ranking read straight off it would have been wrong.

  3. Research

    I profiled the file before deciding anything, counting missing values per column, checking for duplicate rows, and reading the distributions to see which fields the ranking would actually rest on. That pass is what showed the data was not analysis-ready.

  4. Process

    I fixed four steps, clean, prep, model, visualise, and kept them separate so every answer could be traced back to one of them. The faster route was to open the raw file straight in Power BI and chart it; I dropped that, because it produces a confident ranking off dirty rows. Charts that did not answer a line in the brief were cut too.

  5. Result

    A Power BI dashboard reading off the cleaned dataset, with a sales overview, the top three products, and the leading product in each city. The recording below steps through it in the order the brief asks its questions.

  6. Impact

    Every question in the brief is answered on one screen among roughly 300 entries, and the ranking holds up because the cleaning behind it is documented rather than assumed.

  7. Takeaway

    Under competition time pressure the instinct is to jump straight to the visual, and this confirmed the cleaning step is what the whole answer stands on. What I would change: write the profiling pass up as a short data-quality note beside the dashboard, so a reader can see why the numbers are trustworthy instead of taking my word for it.