Rounding, Approximation & Sanity Checks
The biggest misconception about rounding is that it makes your answer “less accurate.” In real business math, rounding is often what makes the answer usable - because 7.4 lakh versus 74 lakh is a strategic difference, while 7,43,219 versus 7,43,220 is noise.
- Rounding means replacing a number with a nearby simpler number that keeps the decision unchanged.
- Approximation means building a reasonable answer from simplified assumptions when exact data is unavailable.
- Sanity checks ask: “Can this answer be true in the real world?” using units, limits, ratios and comparable benchmarks.
- Round inputs early, but keep one guard digit when errors can compound - for example, use 1.2 crore, not exactly 1 crore.
- Use the interview loop: define, round, calculate, sanity-check, refine.
- Never present false precision. “About ₹15 lakh per month” sounds sharper than “₹14,42,880” if assumptions are rough.
- The strongest candidates say the range, the logic and the sanity check - not just the final number.
Big Picture - Rounding Is a Control System, Not a Shortcut
Good estimation is a loop. You simplify numbers to move fast, calculate the answer, test whether it could be true, and then refine only the assumptions that materially change the result.
Core Explanation - The Skill Behind Fast, Credible Answers
Rounding, approximation and sanity checks work together. Rounding makes numbers manageable. Approximation helps you proceed without perfect data. Sanity checks stop you from giving a mathematically neat but commercially absurd answer.
Before you calculate, define the problem clearly. If the question is “estimate the market for premium coffee in Mumbai,” decide whether you mean revenue, number of cups, number of customers or store opportunity. That habit is the natural first step after defining the problem before solving it.
The Five-Step Interview Method
The Rounding Ladder - How Much Precision to Keep
The level of precision depends on the decision. A CEO deciding whether a market is ₹50 crore or ₹500 crore does not need paise-level accuracy. A pricing analyst deciding contribution margin may need tighter numbers.
Definitions You Can Say in One Breath
- Rounding: Replacing a number with a nearby simpler value while preserving its practical meaning.
- Approximation: Estimating an answer using simplified assumptions when exact data is unavailable or unnecessary.
- Sanity check: A quick test that verifies whether an answer is directionally possible in the real world.
- Order of magnitude: The power-of-ten scale of a number, used to detect 10x errors quickly.
The Sanity-Check Toolkit
A sanity check is not a vague “does this feel right?” moment. Use concrete tests. These are the ones that save candidates in guesstimates, profitability cases and market-sizing questions.
For profitability cases, pair sanity checks with contribution logic. If your rounded estimate says a business sells more but loses money faster, the next natural topic is contribution margin and break-even analysis in cases.
Worked Example - Rounding a Café Revenue Estimate
Question: Estimate monthly revenue for a 100-seat café near a B-school.
The final answer should not be “₹14,40,000 exactly.” The assumptions are approximate, so the answer is better stated as: “roughly ₹14-15 lakh monthly revenue, before costs.”
Exact Math vs Interview Math
Exactness is useful when the data is exact and the decision depends on small differences. Interview math is different: it tests structure, judgement and error control under uncertainty.
Case Study - Zepto and the Arithmetic of Quick Commerce
Zepto shows why quick-commerce decisions need ruthless approximation: a small error in orders, basket size, delivery cost or picking capacity can flip the business logic.
Quick commerce is a perfect arena for rounding and sanity checks because the business is dense, local and operational. A dark store must handle enough nearby demand to justify rent, staff, inventory, technology and rider movement. The primary driver is proximity-based fulfilment density: many small orders clustered close enough to be served quickly. Supporting drivers include assortment discipline, inventory replenishment, rider availability, app-led demand generation and tight store processes.
Now imagine evaluating one dark store. You do not begin with a 14-cell spreadsheet. You begin with a rounded unit-economics spine:
The lesson is not “quick commerce works” or “quick commerce fails.” The lesson is that a rounded model exposes the pressure points early. If your estimate needs impossible order density or assumes delivery cost is almost zero, the sanity check catches the flaw before the spreadsheet hides it.
How AI Changes Rounding, Approximation & Sanity Checks
AI makes this topic more important, not less. Tools can calculate instantly, but they can also make weak assumptions look polished. In 2026, the candidate’s edge is knowing what to ask AI to check.
- AI reduces arithmetic load: ChatGPT, Claude or spreadsheet assistants can compute scenarios quickly, so your value shifts to defining the drivers and judging whether the assumptions make sense.
- AI increases false precision risk: An LLM may return a neat-looking number even when the input assumptions are vague. You must force ranges, units and ceilings.
- AI improves challenge practice: You can ask a tool to attack your estimate from the outside: “Find the 3 assumptions most likely to be wrong and give a sanity check for each.”
Use ChatGPT or Claude after solving a guesstimate: paste your structure and ask, “Check units, order of magnitude, capacity limits and penetration ceilings. Do not solve from scratch.” Then practise follow-up questioning through AI as a mock interviewer.
Interview Relevance
“Estimate the monthly revenue of a food court in a large mall. You may use assumptions, but walk me through your rounding and sanity checks.”
When stuck, say: “Let me first get the order of magnitude right, then I’ll refine the two assumptions that matter most.” That sounds structured and buys thinking time.
Common Mistake
The mistake: giving a precise final number from rough assumptions, without a sanity check. It costs candidates because it signals calculator thinking, not business judgement. Fix: present a rounded range and add one unit check plus one real-world ceiling check.