Level 6 · Metric trap: the margin that rose by itself

A blended rate has two moving parts

The store’s gross margin % is not a separate measurement. It is a weighted average of the category margins, each weighted by the category’s share of net sales:

store_margin = (share * margin).sum()   # one row per category

So it can change for two reasons: the rates change (categories get better or worse margins), or the mix changes (sales shift between categories that already had different margins). A report that only shows the blended number cannot tell the two apart.

What happened between Q2 and Q4

Harbor & Pine’s Outdoor category peaks from April to July. In Q2 2025 it was 39.5% of net sales on lines with a known cost; in Q4 it was 9.4%. Outdoor also has the thinnest margin, around 50%, while the other categories sit between 52% and 57%. When the low-margin group shrinks, the average climbs even if nothing else changes.

Meanwhile Q4 was the holiday quarter, with discounts of 3.11% of gross sales against 1.64% in Q2. That pushed four category margins down. The two effects went in opposite directions, and mix won.

Splitting the change

The standard check is to recompute the current period at the old mix:

q4_at_q2_mix = (q2_share * q4_margin).sum()   # 52.77

• Q2 actual: 53.13%
• Q4 at Q2’s mix: 52.77% (rate effect: −0.36 points)
• Q4 actual: 54.01% (mix effect: +1.24 points)

The rates alone made Q4 slightly worse. The mix made it look better.

Why not just average the categories?

The plain average of the six category margins fell from 54.45% to 53.75%. It points the right way here, but it is the wrong number in general: it gives a small category the same weight as a big one. The honest split is the one above, with sales shares as the weights.

Common mistakes

• Thanking or blaming a team for a mix change. Nobody priced better; the season changed.
• Comparing blended rates across periods with different mixes (summer vs holiday, new vs old customers, mobile vs desktop) without the split.
• Using a simple average of group rates as the total.

Interview note

This is Simpson’s paradox in a business setting, and interviewers love it. Say “I would split the change into a rate effect and a mix effect by recomputing at a fixed mix”, and you have answered it.