The brief
A one-month internship task: analyse an Amazon India sales report covering March to June 2022 and turn it into recommendations. The brief set six areas: sales trends, products, fulfilment, customer segments, geography and business actions.
Cleaning decisions
- Two columns were entirely empty and were dropped.
- Amount is blank on 7,800 rows (6.0%), mostly cancellations. Those rows still count as orders but stay out of value totals.
- State names arrived in 69 spellings for about 36 states. Trimming and upper-casing brought that to 47. I left the remaining genuine misspellings unmerged, so fuzzy matching couldn’t join two different places.
- A repeated order ID is a line item, not a duplicate: 128,976 line items across 120,229 orders.
What it found
- Gross order value (₹78.59M) overstates realised revenue (₹70.28M) by ₹8.31M, 10.6% of the gross figure, once cancelled, returned and failed lines are taken out.
A note on duplicates
Revisiting this analysis in November 2026, I found 959 rows that are exact copies of another row. The file has no product-level identifier, so it can’t tell two identical units in one order from an export duplicate. I kept the rows and report every figure as it stands. Removing them would lower the totals by about 0.5 to 0.7 percent; the check is in the repository.