Level 7 · Boss: the order value report, done right

The shape of a good descriptive report

A descriptive slide answers four questions in order: what data (and what was removed), where the middle is, how spread out it is, and how it splits into the groups the business cares about. Each step of this boss is one of those blocks.

Step 1: the cleaning note

From level 5: a negative completed order is impossible, so it is an error. Count what you remove; that count goes into the footnote.

bad = completed["order_value"] <= 0
bad.sum()      # how many to exclude

Step 2: comparing groups with mean and median

Comparing channels by their means alone mixes two things: what a typical customer spends, and how many very large orders the channel happens to get. Put the median next to the mean:

clean.groupby("channel", as_index=False).agg(
    orders=("order_value", "size"),
    mean=("order_value", "mean"),
    median=("order_value", "median"),
)

If the means differ and the medians don’t, the difference lives in the tail. If both differ in the same direction, the whole distribution has shifted. Those are different stories for a sales team.

Step 3: profiling segments

For two very different groups, give each its own middle and spread, and say how much each matters to the business:

g = clean.groupby("segment")["order_value"]
g.median()                                   # middle
g.quantile(0.75) - g.quantile(0.25)          # IQR: spread of the middle half
g.sum() / clean["order_value"].sum()         # share of order value

Look at the two medians before you look at anything else. When one segment’s typical order is three or four times the other’s, a single store-wide average sits between them and describes neither: it is too high for consumers and too low for business buyers. Any campaign, target or forecast built on it will be wrong for both groups.

A share of the total is often the most persuasive single number in a report: “a quarter of our orders bring in half of our revenue” tells a sales leader where to look.

Common mistakes

• Reporting means only. Two groups with different tails can swap places between mean and median.
• Silent cleaning. Every exclusion goes in the notes, with a count.
• Mixing grains. Count orders with size, not customers; one customer can place many orders.
• Shares that don’t add up. Divide each segment’s sum by the total of the same cleaned table, so the shares sum to 1.