When two or more segments each carry their own rate (a margin, a growth rate, a win rate, a price) and you need the combined figure for the whole, you want a weighted average. Weight each rate by the base it applies to, and leave the flat average of the rates alone:
Blended rate = (weight₁ × rate₁ + weight₂ × rate₂ + …) ÷ (weight₁ + weight₂ + …)
Weighted average vs simple average: an example
The usual failure is reaching for the simple average instead, and it's worth seeing how far wrong that goes. A retailer's Category A does ₹60 crore of revenue at a 25% margin, and Category B does ₹40 crore at a 10% margin. The correctly weighted blend is (60 × 0.25 + 40 × 0.10) ÷ 100 = (15 + 4) ÷ 100 = 19%. The flat average of 25% and 10% is 17.5%. That's close enough to feel safe and wrong enough to matter, and the gap only grows as the two segments drift further apart in size or rate. The rule goes well beyond margins. It covers any blended rate, whether that's growth, price, conversion or win rate. Wherever two sub-populations combine into one headline number, weight by the base. The trap is easy to fall into because two rates on a slide invite a quick mental average, and nobody in the room objects until the totals refuse to add up.
Advertisement
What is the mix-shift effect?
A second, subtler failure hides beneath the first. The blend can move when no underlying rate changes at all, purely because the weights shift. That's the mix-shift effect. It's one of the most useful diagnostic distinctions in case work, because it separates "the business got structurally better" from "the mix moved toward the part that was already better." Carry the example forward a year. Category A grows to ₹70 crore, still at a 25% margin, while Category B holds at ₹40 crore, still at 10%. Neither margin has moved a single point. The new blend is (70 × 0.25 + 40 × 0.10) ÷ 110 = (17.5 + 4) ÷ 110 = 19.5%, up half a point only because the higher-margin segment now carries more weight. In a real case, the first thought this should trigger is never "great, margins are improving." Ask "is this mix, or is this like-for-like?" The two answers say very different things about whether the improvement will last.
{
"skill": "Generating a bottomless, leveled stream of case-math drills to build speed reps outside the room",
"use": "Ask an AI chat assistant to generate a batch of percentage, growth, weighted-average, breakeven, and ratio problems at a stated difficulty, mixing India (₹) and US ($) scenarios, so you can drill speed and accuracy on your own schedule — never as a stand-in for doing the arithmetic yourself, live, in the room.",
"tools": ["Chat assistants — Claude, ChatGPT, Gemini", "Spreadsheet tools — Excel or Google Sheets to auto-generate randomized drill batches with a formula-built answer key"],
"prompt": "Generate 15 case-math problems at <difficulty level> covering <percentages / growth-CAGR / weighted averages / breakeven / ratio checks>, mixing India (₹) and US ($) scenarios and using clean, case-realistic numbers. Give me only the problems first, with no answers. On a second message, give me the fully worked answer key, one step at a time, with every intermediate number shown.",
"verify": "Recompute every answer in the key by hand before trusting it — chained, multi-step arithmetic is exactly where a model's token-level, non-exact computation is least reliable, the same failure mode that makes long multiplication shaky for a bare model. Treat any reference magnitude a drill borrows from the real world (an actual price, an actual market figure) as unverified until traced to a named primary source; a fully synthetic drill number needs no external trace but must still reconcile internally — recompute it end to end yourself (Guardrail: Trace, Recompute).",
"diy": "You must be able to solve every generated problem unaided, at speed, narrating as you go, before you ever look at the AI's key — the drill only builds the reflex if you write your own answer first. In the actual case room there is no chat window; the reflex has to already live in your own head and your own hand.",
"market": "US"
}