Revenue Calculations and Feature Engineering

The most fundamental calculation in a sales dataset is line-item revenue: the product of quantity and unit price for each row. Create the column with a vectorised multiplication so NumPy processes every row simultaneously without a Python loop. This column becomes the basis for every aggregation in the analysis.

Grouping by calendar period is essential for trend analysis. Use the .dt accessor on a datetime column to extract year, month, and quarter. Storing these as separate integer columns makes groupby operations faster and lets you filter by period easily without repeated string parsing.

Business rules often define a discount flag: orders where the total exceeds a threshold receive a discount. Use np.where or a boolean comparison to create this binary column efficiently. A discount flag lets analysts compare revenue distributions between discounted and full-price orders in a single groupby call.

If the dataset includes a cost_price column, you can compute profit margin as (unit_price - cost_price) / unit_price. Use clip(lower=0) to cap any negative margins caused by data errors before further analysis. Margin columns enable profit-focused groupby aggregations alongside revenue.

The number of days between a customer's first order and the current order is a useful customer tenure feature. Compute it by grouping on customer_id, taking the minimum date as the cohort date, and subtracting from each row's date. Timedelta arithmetic returns a column of timedelta64 values that you convert to integer days with .dt.days.

Bin the revenue column into labelled size buckets using pd.cut() to create an ordinal feature for segmentation. Provide explicit bins and labels that match your business definitions of small, medium, and large orders. The resulting column is a Pandas Categorical, which groups efficiently.

Sort the DataFrame by order_date and use cumsum() on the revenue column to track how total revenue accumulated over the year. This cumulative series makes it easy to spot whether revenue is growing linearly, accelerating, or plateauing, and is the basis for a waterfall or cumulative line chart.

The dt.day_of_week property returns 0 (Monday) through 6 (Sunday). Flag Saturday (5) and Sunday (6) as weekend orders to test whether weekend buying behaviour differs from weekday. This is a common feature in retail EDA and A/B test analysis.

Identify your top customers by computing total revenue per customer_id with groupby().sum() and then ranking with .rank(ascending=False, method='dense'). Adding the rank back to the main DataFrame with map() lets you filter to top-N customers in a single boolean condition.

Normalise the revenue column to a Z-score so you can compare order sizes across different product categories that have very different price ranges. Use (df['revenue'] - df['revenue'].mean()) / df['revenue'].std(). Z-scores above 3 or below -3 are statistical outliers worth investigating before modelling.

After adding computed columns, save the enriched DataFrame so every downstream notebook starts from the same consistent state. Use to_parquet() to preserve dtypes, especially the Categorical order_size column and the datetime order_date. Document the new columns in a short comment block at the top of the script.

Test your understanding of Data Analysis concepts from this lesson.

In this lesson you learned: computing revenue and profit margin columns, extracting temporal features with the .dt accessor, and engineering features like discount flags, order size buckets, and customer ranks. Next up we explore groupby analysis by region and category.