• Raw data extraction from KAMIS’s live market search tool (7 commodities × 11 counties)
• Power Query ETL pipeline — combining, cleaning, and transforming ~10,800 raw price records
• A proper star-schema data model in Excel’s Data Model (Power Pivot), not just a flat spreadsheet
• DAX measures for volatility, margins, and data-quality checks
• An interactive dashboard with slicers for County, Region, and Year
• A full findings & recommendations report, written for both technical and non-technical readers

Real data doesn’t arrive clean, and this project leaned into that instead of hiding it. A few things I found and had to solve along the way:

• Some counties simply weren’t in the system. I originally planned to cover Kiambu, Machakos, Mombasa, and Kajiado — but testing showed zero price records for any of them, across every commodity. Rather than force it, I swapped in better-covered counties and documented why.
• One market was quietly broken. Molo (in Nakuru county) never once recorded a wholesale price, and its retail prices swung wildly with no coherent pattern. I traced it, confirmed it was a structural reporting issue rather than a one-off typo, and excluded it — with a clear note explaining why.
• A few prices were off by a factor of 90. Some entries looked like someone had typed a price-per-90kg-bag into a price-per-kilogram field. I found these by comparing wholesale and retail prices against each other and flagging implausible ratios, not just eyeballing the data.
• Small samples were quietly lying. A couple of county-commodity pairs looked “wildly volatile” — until I checked and found they only had 5 or 6 data points behind them. I added a minimum sample-size rule before trusting any volatility number.

None of this was hidden after the fact — every decision is logged in the report’s methodology section, including exactly how many rows were affected and why.