Level 9 · Boss: the channel report for the board

Two grids, one meeting

This boss uses the chapter’s grid tools on a real board request. The data starts long: one row per order line or per order. The board reads grids, and asks questions about them.

Each step runs on a fresh copy of the tables, so each one builds what it needs again:

items = order_items.merge(orders[["order_id", "order_date", "status", "channel"]], on="order_id")
items = items[items["status"] == "completed"].copy()
items["revenue"] = items["quantity"] * items["unit_price"] * (1 - items["discount_pct"] / 100)
items["year"] = items["order_date"].dt.year

Growth in wide form

With channel down the side and year across, completed revenue totals 1,600,511.41 for 2023, 1,993,605.32 for 2024 and 2,300,170.84 for 2025. Each year is a column, so growth is column arithmetic, computed for every channel at once:

grid["growth_2025"] = grid[2025] / grid[2024] - 1

The labels came from dt.year, so they’re integers. Compute growth on the unrounded totals, then round. Sorted by growth, the story is plain: the app grew 53.1% and became the biggest channel, the marketplace grew 13.3%, web fell 7.2%.

Months in calendar order

dt.month_name() turns a date into “January”, “February” and so on, which is what a reader wants to see. The catch: pandas sorts text alphabetically, so a pivot on month names comes out April, August, December, February… Two fixes:

months = ["January", "February", "March", "April", "May", "June", "July",
          "August", "September", "October", "November", "December"]
grid.reindex(months)                     # rows in the order of the list

or keep the month number (dt.month) next to the name, sort by the number, and then set the name as the index. reindex is shorter when you already have the list; it also adds a row of NaN for any month that’s missing, which makes gaps visible instead of silently dropping them.

Reading the grid: idxmax and sum(axis=1)

Two short methods answer most “which one?” questions:

• .idxmax() on a Series returns the label of its largest value (.max() returns the value). grid.loc["January"].idxmax() is the channel that sold most in January.
• grid.sum(axis=1) adds across each row (one total per month); grid.sum() adds down each column (one total per channel). Chain .idxmax() to name the biggest.

grid.idxmax(axis=1)        # the leading channel in every month

In 2025 the app led nine months, but web led January, June and August. A headline like “the app leads every month” would be wrong, and these two lines are how you check it.

Common mistakes

• Pricing lines with products["unit_price"] instead of the price on the order line.
• Writing grid["2025"] for an integer column label.
• Leaving months in alphabetical order.
• Using .max() when the question asks which month or channel: that’s .idxmax().
• Summing with axis=0 when you want one total per row.