Guesstimate Interview Questions: The Method, Worked Examples, and a Practice Set

Why interviewers ask guesstimates at all

A guesstimate is a proxy for the job. Analysts spend their days sizing things nobody has measured: how large a market is, whether a metric spike is plausible, what a feature might earn. The interviewer already knows you cannot know the answer. What they are watching is whether you panic or structure, whether your assumptions are stated or smuggled, and whether the arithmetic you commit to actually holds together. That changes how you should prepare. Memorising answers is close to useless, because the numbers are not the point and the question bank is infinite. What transfers between every guesstimate is a method, and the method fits in five steps: clarify the question, choose a structure, segment the population, estimate each segment with stated assumptions, and sanity-check the total from a second direction. A candidate who runs those five steps calmly on a question they have never seen beats one who recites a rehearsed Fermi estimate every time. One more thing worth knowing before the examples: interviewers routinely push on one of your assumptions midway. That is not a trap closing. It is the best part of your answer, because adjusting one input and re-deriving the total is exactly the sensitivity thinking the job requires.

The five-step method

Clarify first. Twenty seconds of questions saves five minutes of estimating the wrong thing. For a delivery question: which city, orders or revenue, one platform or the whole market, an average day or a peak. Choose a direction second. Demand-side starts from people and works down through adoption and frequency. Supply-side starts from capacity: riders, ATMs, seats, servers. Pick whichever has assumptions you can defend, and remember the other one, because it becomes your sanity check at the end. Segment third. An average across a whole population hides the structure that makes your estimate defensible. Splitting a city into users and non-users, or heavy and light orderers, turns one impossible guess into three easy ones. Estimate fourth, with round numbers. Interviewers grade clean arithmetic under pressure, so 13 million beats 12.8 million and 40 percent beats 37 percent. Say each assumption as you use it, so the interviewer can challenge the input rather than the output. Sanity-check last, from a different angle. If demand-side gave you the number, spend thirty seconds on supply-side. When the two disagree, and they often will, name which assumption drives the gap. That single sentence is worth more than either estimate.

Worked example: food delivery orders per day in Bengaluru

Clarify: all platforms combined, a normal weekday, orders rather than revenue. Demand side. Assume Bengaluru holds about 13 million people. Assume 40 percent are active food delivery users, reflecting a young, smartphone-heavy city: that is roughly 5 million users. Assume the average active user orders twice a week, which blends daily orderers with occasional ones: 5 million times 2 divided by 7 gives about 1.5 million orders a day. Sanity check from the supply side. A delivery rider completes perhaps 15 to 20 orders in a day. Serving 1.5 million orders would need roughly 75,000 to 100,000 riders active daily across all platforms in one city. For a city Bengaluru's size that is plausible, if toward the high end, so the estimate stands with a stated range: around 1 to 1.5 million orders a day. Notice what did the work. Every input was a stated assumption an interviewer could push on. If they challenge the 40 percent adoption, you drop it to 30, re-derive 1.1 million in one line of arithmetic, and the structure survives. That resilience under pushback is the pass signal.

Worked example: how many ATMs does a city of 5 million need?

This one is chosen because the two directions disagree, which is the situation that separates strong candidates. Demand side. Of 5 million people, take 70 percent as adults: 3.5 million. Assume the average adult makes 3 ATM withdrawals a month, which is about 350,000 transactions a day across the city. A single ATM can serve perhaps 120 transactions a day comfortably. That suggests roughly 3,000 ATMs. Anchor check. India operates on the order of 250,000 ATMs for about 1.4 billion people, which is one ATM per 5,500 to 6,000 people. Scaling that ratio, a 5 million person city would hold about 900 ATMs. The two answers differ by three times. The wrong move is to quietly pick one. The strong move is to explain the gap: the national ratio includes villages with sparse coverage, while a city has denser usage, and the transactions-per-adult assumption drives most of the difference in a digital-payments era where cash withdrawal frequency is falling. So the defensible answer is a bracket, roughly 1,000 to 3,000, with urban density arguing for the upper half. Naming which assumption moves the answer is exactly the sensitivity analysis the role demands.

The practice set: 35 questions by type

These are from our curated interview bank, grouped by the muscle they train. Do not read answers for these. Set a ten-minute timer, run the five steps out loud, then audit your own assumptions. Consumer scale: food delivery orders in Bengaluru per day, daily active users of a ride-hailing app in Mumbai, smartphones sold in India in a year, metro tickets on a weekday, flights out of Delhi airport in a day, weekly demand for a subscription box, online grocery market size in a tier-2 city. Business and revenue: annual revenue of one mall coffee outlet, sizing the addressable market for a B2B analytics tool, estimating a competitor's revenue from public data only, the revenue impact of a one-day outage, laptops a large IT company replaces in a year. Infrastructure and operations: ATMs for a city of 5 million, petrol pumps in your state, delivery riders needed in one zone, support agents for a 10 million user app, UPI transactions per second at peak, streaming data served per day in India, storage cost of a year of clickstream. Probability and puzzles: the shared-birthday probability in a room of 30, the false-positive test question, two aces from a deck, the expected rolls to see a six, the coin-toss game that pays 10 and costs 4, the mislabelled jars, 100 toggled doors, the two unevenly burning ropes, 8 balls and two weighings, a stopping rule for interviewing 100 candidates, detecting a biased coin efficiently. Judgment under ambiguity: a metric that doubled overnight, what revenue and order count alone can and cannot tell you, two conflicting reports with no time, and how you sanity-check any number someone hands you. That last one is quietly the most important question on this page, because it is the job.

How you are graded, and the mistakes that fail candidates

Interviewers running guesstimates score four things: structure, stated assumptions, arithmetic under pressure, and the sanity check. A wrong final number with clean structure passes. A lucky number with hidden assumptions does not. The failure patterns repeat across candidates. Diving into numbers before clarifying scope, so the whole answer sizes the wrong thing. Smuggling assumptions inside multiplication instead of stating them, which reads as either carelessness or evasion. False precision, quoting 1,347,000 where 1.3 million was available, which signals discomfort with estimation itself. Skipping the sanity check, which is the single most commonly dropped step and the cheapest to keep. And treating a pushback as an attack, defending the original number instead of happily re-deriving with the new input. One habit fixes most of these: narrate. Say what you are about to do, do it, say what you got, and say what would change it. Silence is where guesstimates die, because the interviewer cannot grade reasoning they never heard.

Are guesstimate answers supposed to be exact?

No, and chasing exactness actively hurts you. The interviewer knows the true number is unknowable in the room. What earns marks is a defensible structure, stated assumptions, clean round-number arithmetic, and a sanity check from a second direction. Quote ranges, not five-digit precision: a candidate who says one to one and a half million with reasons beats one who says 1,347,000 with confidence.

How long should a guesstimate answer take?

Eight to twelve minutes is the common window for a full walk-through: about a minute clarifying, a minute choosing structure, five or six estimating out loud, and a genuine sanity check at the end. Rushing to a number in two minutes reads as rehearsed; wandering past fifteen reads as unstructured. Practising with a ten-minute timer builds the right internal clock.

Which roles actually get guesstimate questions?

Data analyst and business analyst screens use them heavily, consulting interviews are built around them, and product analyst and product manager loops include them routinely. Data engineering interviews use them less, though capacity questions such as storage costs or transactions per second are the same skill wearing infrastructure clothes. If your target is an analyst role in India, assume at least one round includes an estimation exercise.

What numbers should I memorise before an interview?

A short anchor list covers most questions asked in India: the country's population, your city's approximate population, the rough split of urban and rural, seconds in a day, and one or two capacity figures such as how many deliveries a rider completes daily. Five to ten anchors are enough. The skill being tested is deriving the unknown from the known, not carrying an almanac in your head.

How are the puzzle questions different from estimation questions?

Puzzles such as the mislabelled jars, the 100 doors, or the two burning ropes have exact correct answers, so the estimation method does not apply. They test structured logic instead: enumerate the cases, find the invariant, work backwards from the goal. Treat them as a separate practice track, and if you are asked one, slow down; the classic mistake is pattern-matching to a puzzle you memorised rather than the one actually asked.

What if my final number is way off?

Say so yourself before the interviewer does. Finishing with a sanity check that flags your own answer as high, and naming which assumption likely inflated it, converts an off number into a demonstration of exactly the judgment the role needs. Candidates fail guesstimates by defending shaky numbers, almost never by catching their own.