Ayush Ummadi

2,823 line items across 307 orders, worked through with CTEs and window functions: revenue trends, product lines, customer behaviour and order operations.

November is the highest-revenue month in every year of the data, and Classic Cars alone account for $3.92M of the $10.03M total.

RFM scoring across 247 transactions, sorting customers into Champions, Loyal, At Risk, New and Lost, then profiling each segment by demographics and product category.

At Risk customers hold the largest revenue pool at $7,830, and they haven't bought in an average of 253 days.

Sales by state on a choropleth, sales against profit over time, discount and profit distributions, and category donuts, covering 2014 to 2017.

Most revenue comes from orders with little or no discount. The heavily discounted tail contributes far less.

Five days of sales against target, built as linked pivot tables and charts that all re-cut together from a single region slicer.

The average target hit rate is 55.2%, with individuals running from 77.8% down to 28.6%.

Excel functions, pivot tables and a demand forecast across product performance, profitability and seasonal demand.

One customer is 32% of revenue, and the top two together are 57% of it.

Gender, age, marital status, occupation and city category tested against what people actually spend.

Unmarried men aged 26 to 35 are the largest segment in almost every cut, and men account for roughly 77% of total purchase value.

A growth metric derived from the raw figures across 256 countries and regions, plus a reusable function for comparing any set of countries.

Only 120 of the 256 have complete coverage across the period, and the sparse ones throw out wildly overstated growth rates.

Production volume, land use and yield per hectare, compared across countries and continents.

Production tracks acreage almost perfectly, a correlation of 0.998, while yield per hectare barely relates to total output at 0.13.

Distributions, violin plots, a correlation heatmap and a pairplot over the UCI heart disease dataset.

172 of the 918 cholesterol readings are recorded as zero, which is missing data rather than a real measurement.