Level 2 · Monthly revenue from order lines

Revenue needs two tables

Counting orders per month only needed orders. Revenue is different: the money is on order_items (quantity, price paid and discount per line), while the date and the status are on orders. So a monthly revenue series always starts by putting those together:

items = order_items.assign(
    line_revenue=order_items["quantity"]
    * order_items["unit_price"]
    * (1 - order_items["discount_pct"] / 100)
)
completed = orders.loc[orders["status"] == "completed", ["order_id", "order_date"]]
lines = items.merge(completed, on="order_id")

Selecting only order_id and order_date from orders keeps the merged table narrow. Filtering to completed orders before the merge means the inner join drops every line of a cancelled, refunded or pending order in one go.

Resample one column

With the date on every line, the pattern from the previous level applies. Select the column you want to aggregate after resample, just as you would after groupby:

monthly = lines.set_index("order_date").resample("MS")["line_revenue"].sum()
monthly.round(2).head(3)
# order_date
# 2023-01-01    85437.41
# 2023-02-01    66624.99
# 2023-03-01    87788.05

Without selecting a column, .sum() would try to add every numeric column, including order_item_id and quantity, which is wasted work and a confusing table.

You can also skip set_index and pass the date column with on=:

lines.resample("MS", on="order_date")["line_revenue"].sum()

Both give the same series. on= is handy when you want to keep the original index.

Reading the series

Completed revenue was about $1.60 million in 2023, $1.99 million in 2024 and $2.30 million in 2025. The best month is November 2024 at $319,217.26, ahead of November 2025 at $292,364.40, even though November 2025 had more orders. A monthly series shows that kind of surprise immediately; the next levels give you tools (rolling averages and year-over-year growth) to measure it.

Rounding at the end

Round after summing, not before. Rounding each line to cents first and then summing gives a slightly different total, and it gets worse as the number of lines grows. Keep full precision through the calculation and round once, for display.

Common mistakes

• Forgetting the status filter, so cancelled and refunded orders inflate every month.
• Resampling order_items directly: it has no date column, so there is nothing to put in the index.
• Merging the full orders table and summing every numeric column after resample.
• Using "ME", which labels each month with its last day when the ledger uses the first.