Pressure-Testing the Answer
The fifteen seconds that catch a 10× error — and the traps that cause them.
A candidate finishes a clean, well-narrated solve and lands on ₹4,000 cr for premium dark chocolate. The structure scored. The arithmetic scored. Then one question — isn't that a quarter of India's entire chocolate market? — vaporises all of it. The tragedy is that a fifteen-second cross-check would have caught it. Triangulation and the sanity check are the cheapest insurance policy in the entire case: together worth 20% of the rubric, costing almost no time, and skipped by most candidates. This page makes them reflex.
TL;DR · Key Takeaways
After this page, you will be able to
- Triangulate every solve with an independent second method built from different anchors, and reconcile the two out loud.
- Run the share, per-capita, and magnitude tests in seconds to catch a 10× error before the interviewer does.
- Treat a large gap between two methods as a diagnostic to debug, never as something to average.
- Close on a central estimate with a range driven by a named swing assumption, tied back to the business 'so what'.
- Pre-flight the eight common traps — slipped zeros, unit drift, flow-vs-stock, over-segmentation, and the rest — as a final checklist.
Triangulate: re-derive from the diagonal
Triangulation is one rule: solve it a second way and reconcile. Page 2 gave you the map — if your first pass was Top-Down on the demand side, your second pass is Bottom-Up or supply-side. The casebook formulation is exactly this: check a top-down answer with a bottom-up one, or vice versa. You don't need a perfect second method; you need an independent one, built from different anchors, so the two errors don't cancel and hide.
Sanity check: does the number make sense?
Before you say the number out loud, hold it against something you already know to be true. Make sure your numbers make sense in the context of the case — the simplest, most-skipped tip in every casebook. Three quick tests catch almost every blunder: the share test (is this a believable slice of a bigger total you know?), the per-capita test (does it imply something absurd per person or per day?), and the magnitude test (count the zeroes — are you in lakhs, crores, or lakh-crores?).
Spot the order-of-magnitude error
For each claimed answer, decide in five seconds whether it's plausible — and if not, which test it fails. Then reveal.
"Premium dark chocolate in India is a ₹4,000 cr / year market."
"There are about 8 lakh petrol pumps in India."
"India does roughly 50 crore UPI transactions per day."
"About 10 crore weddings happen in India each year."
Recommend: a number with bounds, and a 'so what'
Don't end on a bare point estimate. End on a central number plus a range driven by your swing assumption, then tie it back to why anyone asked. "≈ ₹900 cr, range ₹500–1,700 cr depending on the active-buyer ratio — so for a brand deciding whether to enter, even the floor clears the ₹300 cr threshold that usually justifies a launch." The range shows you know your estimate is a model, not a fact; the 'so what' shows you remember it was a business question, not an arithmetic one.
Name your swing assumption — then bound it
Every guesstimate has one assumption that moves the answer more than all the others combined. Find it, say it, and flex it: "if the ownership rate is 8% instead of 12%, the answer drops to ~4M; at 16% it's ~8M." A bounded answer with a named driver reads as senior; a single number read off the top of a calculation reads as junior.
The trap list
Most lost points cluster into a handful of repeat offenders. Learn them as a pre-flight checklist — a glance down this list in the last ten seconds catches the error that ends rounds.
| Trap | What it looks like | The fix |
|---|---|---|
| Slipped zero | Answer is 10× or 100× off; lakhs muddled with crores. | Work in powers of ten; say the magnitude aloud at each step. |
| Unit drift | Per-day mixed with per-year; litres with bottles. | Lock units in Clarify; write them next to every number. |
| Flow vs stock | Counting the installed base when asked for annual sales. | Re-read the prompt: is it 'how many exist' or 'how many sold'? |
| Over-segmentation | Six filters, each ±50%, compounding into noise. | Stop at 3–4 splits; more precision is false precision. |
| Borrowed anchor | A half-remembered statistic used as bedrock. | Derive shaky numbers; only anchor on what you'd defend. |
| Blind averaging | Two methods 5× apart, answer = the mean. | A gap is a diagnostic — fix the broken term, don't split it. |
| Forgotten utilisation | Assuming every asset runs at full capacity. | Apply a utilisation/idle factor to supply-side counts. |
| No 'so what' | A number with no link to the business question. | Close by tying the estimate to the decision it informs. |
Where this connects
This closes the method spine: Page 1 framed the rubric, Page 2 gave the four approaches, Page 3 trained the clarify-and-communicate flow, and this page drilled the triangulation, sanity check, and traps that protect your score. Next, the India cheat sheet (Page 5) hands you the anchors these checks lean on — population, households, incomes, the numbers worth memorising. Then the worked examples in this section let you run the full spine end-to-end; each one tags the approach and the swing assumption so you can pressure-test your own pass against it.