Practice free

Guesstimate and Case Interview Questions for Freshers (2026) — Worked Answers

Updated August 2026

If you are interviewing at a professional-services or consulting firm — our EY, KPMG and PwC guides cover the Indian intakes — you will meet a question with no correct answer available to you. How many two-wheelers are there in Hyderabad. Why have this retailer's sales fallen. Estimate the number of haircuts in this city today. Candidates find these disproportionately frightening, and the fear comes from a misunderstanding about what is being measured.

You are not being tested on whether you know the number. The interviewer usually does not know it either. You are being tested on whether you can take a vague problem, break it into parts that can be reasoned about, state what you are assuming, do arithmetic without panicking, and notice when your own answer is absurd. That is a genuine professional skill — it is what an analyst does on the first day with incomplete client data — and unlike most interview skills it is learnable in a couple of weeks.

This set is written for freshers without an MBA casebook, because you do not need one and using half-remembered frameworks badly is worse than using none. What you need is a method, a handful of anchor numbers, four or five worked examples in your head, and the habit of thinking aloud. Everything below is worked through with real arithmetic rather than described in the abstract, because watching the reasoning is how this is learned.

Frequently asked questions

What is a guesstimate, and why do interviewers ask them?

A guesstimate is a question asking you to estimate a quantity you cannot look up, using reasoning from things you do roughly know. They are asked because they simulate the actual job: a client asks how big an opportunity is, nobody has the data, and somebody has to produce a defensible number by tomorrow. What the interviewer learns from watching you is whether you can decompose a problem, whether your assumptions are sensible, whether you can do arithmetic under mild pressure, and — most revealing — whether you notice when your answer is ridiculous. They also learn how you behave when you are uncomfortable, which is genuinely part of the assessment. Say some of that back if you are asked why they ask: candidates who understand the purpose approach the question very differently from ones who think it is a trivia test.

What are they actually scoring?

Four things, in roughly this order of weight. Structure: did you break the problem into parts that do not overlap and together cover the whole, rather than grabbing at numbers. Assumptions: did you state them out loud, are they reasonable, and did you flag which ones your answer is most sensitive to. Arithmetic: can you multiply and divide cleanly, keep track of zeros, and stay in round numbers. And sanity: did you look at your own answer and ask whether it is plausible, ideally with a cross-check from a different direction. The final number carries almost no weight — being within an order of magnitude is fine, and interviewers frequently accept answers they believe are wrong because the reasoning was sound. What fails is silence, false precision and an unexamined absurd result.

What is the basic method?

Five steps, and following them visibly is most of the marks. First, clarify: repeat the question in your own words and check scope — the city or the country, per day or per year, all types or one. Second, structure: say out loud how you will break it down, ideally in one sentence, so the interviewer knows where you are going before you start calculating. Third, assume: put numbers on each piece, saying where each comes from, and use round numbers because you are doing this mentally. Fourth, calculate: multiply through carefully and keep zeros visible. Fifth, sanity-check: state the answer, then test it from another angle or against something you know, and say what would change it most. The whole thing should take five to ten minutes and be narrated continuously. A candidate who thinks silently for three minutes and produces a correct number scores worse than one who talks through a rougher answer.

Top-down or bottom-up — how do I choose?

Top-down starts from a large known population and narrows it with proportions: start with the population of a city, get to households, then to households owning the thing. Bottom-up builds from a unit and multiplies up: one shop sells this many per day, there are this many shops. Use top-down when the quantity is driven by people — ownership, consumption, subscriptions — because population is the number you actually know. Use bottom-up when it is driven by capacity or supply — how many haircuts, how many flights, how many meals served — because the constraint is the number of chairs, runways or tables rather than desire. The strongest answers use one to estimate and the other to check, and saying "let me sanity-check that from the supply side" is one of the single most impressive moves available to a fresher in this format.

Which anchor numbers should I memorise?

A small set, held loosely and stated as approximations. India's population is roughly 1.4 billion. A large Indian metro is roughly 1 crore, or 10 million — Hyderabad, Bengaluru, Chennai are all in that region; Mumbai and Delhi larger; a tier-2 city like Vijayawada or Visakhapatnam is closer to 1 to 2 million. Average household size is about 4. Roughly a third of India is urban. Life expectancy is around 70, which gives you a rough age distribution if you assume even spread. A working day is 8 to 10 hours; a shop might be open 12. A year is 365 days, but about 300 working days. Do not memorise more than that. The point of an anchor is to give you a defensible starting point, and stating one as "roughly a crore, let us use ten million for easy arithmetic" is exactly right — quoting a precise-sounding figure you cannot defend is worse than a round one you can.

Worked example: how many two-wheelers are there in Hyderabad?

Population-driven, so top-down. "I will take Hyderabad as roughly one crore people, ten million for arithmetic. With an average household of four, that is about 2.5 million households. Two-wheeler ownership in an Indian metro is high but not universal — many households use public transport or own a car instead — so I will assume about 60 per cent of households own at least one, which is 1.5 million households. Some own two, so let me use an average of 1.2 per owning household, giving about 1.8 million. Then add commercial use — delivery riders, small businesses — which I will put at ten per cent on top, so roughly 2 million two-wheelers." Then sanity-check: "That is one two-wheeler for every five people, which feels plausible for an Indian metro. If ownership were nearer 80 per cent of households my answer would rise to about 2.5 million, so that assumption is the one driving the result." Notice that the last sentence — naming the sensitive assumption — is what separates a good answer from an adequate one.

Worked example: how many cups of tea are sold near a college gate in a day?

Small and local, so this is the one to do from both sides. Supply side: "I will assume five tea stalls around the gate. A busy stall serves steadily through the morning and evening peaks and slowly between, so perhaps 300 to 400 cups a day. Taking 350, five stalls give about 1,750 cups." Demand side as the cross-check: "The college has around 3,000 students plus a few hundred staff. Perhaps 40 per cent buy tea on a given day, which is 1,200 people, and those who do average 1.5 cups, giving about 1,800." Then close: "The two methods land at roughly 1,750 and 1,800, so I am reasonably confident the answer is around 1,800 cups a day." That agreement is worth saying out loud, because it demonstrates the check rather than just performing it — and when the two sides disagree by a lot, saying so and working out which assumption is wrong is equally strong.

Worked example: how many haircuts happen in a city in a day?

Capacity-driven, so bottom-up from the supply side. "Take a city of ten million. Salons and barbers are dense in India — I will assume one outlet per 2,000 people, giving 5,000 outlets. A typical outlet has three chairs and operates about ten hours. A haircut takes roughly 30 minutes, so a chair could do 20 a day at full occupancy, but nobody runs at full occupancy, so I will assume 50 per cent, giving 10 per chair, 30 per outlet. Five thousand outlets times 30 is 150,000 haircuts a day." Cross-check from demand: "Ten million people; men cut their hair perhaps every six weeks and women far less often, so across the whole population an average interval of around 60 days seems reasonable. Ten million divided by 60 gives about 167,000 a day." Then: "150,000 and 167,000 are close enough that I would answer roughly 150,000 to 170,000." Two independent routes converging is the strongest possible finish to a guesstimate.

Worked example: how many laptops are there in a large IT campus?

Headcount-driven, and the value here is in remembering the non-obvious categories. "Assume the campus holds 10,000 employees. Essentially every employee in an IT services or product company has a work laptop, so that is 10,000 as the base. Add unassigned stock — spares, machines held by IT for replacements and new joiners — at perhaps five per cent, so 500. Add machines used for testing, builds or demonstration setups, which vary hugely but let me put at another two per cent, 200. Some employees also carry personal laptops onto campus, but I will exclude those as the question is about the campus rather than about people." That gives roughly 10,700, so I would answer about 10,000 to 11,000." The thing being tested here is completeness rather than arithmetic — a candidate who answers "10,000, one per person" and stops has not been wrong, but has shown less. Stating what you deliberately excluded, and why, is part of the answer.

Worked example: how many flights take off from a major airport in a day?

Constraint-driven, and the useful insight is to find the binding constraint rather than counting demand. "The constraint at an airport is runway capacity. A single runway can handle roughly one movement — a takeoff or landing — every two minutes at busy periods, so about 30 an hour. Assume the airport operates at high intensity for around 16 hours and lightly overnight; call it an effective 18 hours at an average of 25 movements an hour, giving about 450 movements. Half of those are takeoffs, so roughly 225 departures a day from one runway. If the airport has two operational runways but they cannot be used fully independently, I would not double it — I would apply perhaps 1.6 times, giving around 350 departures." Then sanity-check: "At an average of 150 passengers a flight, that is over 50,000 departing passengers a day, or nearly 20 million a year, which is in the right region for a large Indian airport." Finding the constraint, and refusing to naively double for the second runway, are both things interviewers notice.

What if the question has no data at all — how many tennis balls fit in this room?

Then it is a pure structure test, and the method is volume divided by volume with a packing correction. "This room looks about 6 metres by 5 by 3, so 90 cubic metres. A tennis ball is about 7 centimetres across, so its bounding cube is roughly 340 cubic centimetres, or 0.00034 cubic metres. Dividing gives about 265,000 if they packed perfectly, but spheres do not — random packing leaves roughly a third of the space empty, so I will take about 65 per cent, giving around 170,000 balls." Then a caveat that shows judgement: "That assumes the room is empty; furniture would reduce it." The mistakes to avoid are guessing a number without a method, and getting lost in precision about packing density — the interviewer wants the structure and one sensible correction, not sphere-packing theory. Physical-estimation questions like this are also where stating your unit conversions out loud protects you from a factor-of-a-thousand error.

How do I state assumptions without sounding like I am making things up?

By giving a reason and a range, and by flagging sensitivity. Compare "let us say 60 per cent of households own a two-wheeler" with "two-wheeler ownership in a metro is high but not universal, since many households use public transport or own a car, so I will take 60 per cent — it is probably between 50 and 70". The second is the same number with reasoning attached, and it invites the interviewer to correct you, which is a good thing rather than a failure. Then, at the end, say which assumption matters most: "my answer moves most with the ownership rate; the commercial add-on barely changes it". That single sentence demonstrates you understand your own model. And if you genuinely do not know something, say so and pick a defensible value rather than pretending — "I do not know the real figure, so I will assume X and flag it" is a perfectly professional sentence.

What do I do if the interviewer challenges a number mid-way?

Treat it as information rather than as an attack, because it usually is one. Often they are testing whether you will defend a reasonable assumption or collapse at the first push. So first, understand what they are objecting to; then either explain your reasoning once, briefly, or accept the correction and say what it does to your answer. Both are good outcomes: "you are right that ownership is probably higher — at 75 per cent my answer moves from about 2 million to about 2.5 million" is an excellent response, because it shows you can update without losing the thread. What scores badly is abandoning your whole structure and starting again, or arguing at length about a number that does not much affect the result. If the challenge is to your structure rather than a number, that is more serious and worth listening to carefully — it usually means you have missed a segment or double-counted one.

Should I write things down or do it in my head?

Write, and write where the interviewer can see it if you are in person. A guesstimate has four or five quantities and one chain of multiplication, and holding those in your head while also narrating is what causes candidates to lose a zero and panic. Write the structure first as a short list or a simple tree, then fill in numbers beside each. It slows you down slightly and it makes your reasoning visible, which is the thing being assessed. In a virtual interview, keep paper beside you and say "let me note that down" rather than going silent. And keep your numbers round — 10 million rather than 9.7 million — because the precision is fake anyway and round numbers are what make mental arithmetic reliable under pressure.

What is a case interview at fresher level, and do I need frameworks?

At fresher level a case is a short business problem discussed conversationally — profits are falling, should we open in a new city, a client wants to cut costs — usually ten to fifteen minutes rather than the elaborate exercise MBA candidates prepare for. You are expected to structure the problem, ask sensible clarifying questions, and reason towards a recommendation, not to produce the answer a consultant would after three weeks. On frameworks: you do not need a casebook, and half-remembered frameworks applied mechanically are actively worse than none — reciting Porter's five forces at a question about a falling profit margin signals memorisation without understanding. What you do need is one genuinely useful decomposition, which is that profit equals revenue minus cost, revenue equals volume times price, and each of those breaks down further. Almost every fresher-level case can be entered through that door.

Worked case: a retail chain says profits have fallen over two quarters. How do you approach it?

Start by clarifying, then decompose, then narrow. "Before I start — has revenue fallen, have costs risen, or both? And is this across all stores or concentrated in some?" Suppose the answer is that revenue is flat and costs have risen. "Then I will focus on cost. Costs split into cost of goods sold and operating costs; operating costs into rent, staff, utilities, logistics and marketing. Which of those has moved?" Say staff costs. "Then I would ask whether that is because headcount rose, because wages rose, or because overtime rose — and whether it is uniform across stores or concentrated in a few." Then a hypothesis: "If it is concentrated in a few stores with rising overtime, I would look at whether those stores are understaffed on the shift rota and paying overtime to cover, which is usually more expensive than hiring." Notice the shape: clarify, decompose into parts that do not overlap, ask which branch to go down, and form a testable hypothesis. That is the whole skill, and it is more valuable than any framework.

Worked case: sales at one branch dropped 20 per cent last month while others were flat. What could explain it?

Being localised to one branch is the key fact, so structure around internal versus external causes for that branch. "Internally: did something change at that branch — a staff departure, a manager change, a stock-out of a popular line, a renovation, changed opening hours, a price or promotion difference? Externally: did something change around it — a new competitor opening nearby, road works or a diversion affecting footfall, a large local employer closing or shifting, a seasonal or event effect specific to that area, or a change in public transport?" Then prioritise rather than listing forever: "The fastest checks are internal and cheap — I would look at stock availability and staffing for that month first, since those are in our own data, before investigating the neighbourhood." And name what would confirm it: "If footfall held steady but conversion dropped, it points internally; if footfall itself fell, it points to something outside the store." Ending with the diagnostic that distinguishes your hypotheses is what makes this a strong answer.

Worked case: should a company launch a product in a new city?

Structure it as a small number of questions that together decide it, and say them before answering any. "I would want to know four things. Is there demand — how many potential customers are there in that city and what is the size of the opportunity, which is a guesstimate in itself. Can we serve it — do we have the supply chain, staff and logistics, or would we be building from scratch. What is the competition — who is already there, and what would make a customer choose us. And what does it cost to enter versus what we expect to earn, and over what period." Then commit: "If I had to decide with no more information, I would want the market size to be large enough that a small share still covers entry costs within a reasonable period, and I would look for a reason customers would switch that is not just price." Two things score here: laying out the structure before diving in, and being willing to state a recommendation with the assumptions it rests on. Refusing to conclude is a common fresher failure.

What if I reach a wrong conclusion?

Reaching a defensible conclusion from stated assumptions is the goal, and interviewers accept conclusions they disagree with when the path is sound. What they cannot accept is a conclusion with no visible path, or one that contradicts your own analysis — a candidate who spends ten minutes showing the market is small and then recommends entering it has not been listening to themselves. So state the recommendation, state what it rests on, and state what would change your mind: "I would enter, but that depends mainly on the entry cost being recoverable within two years; if the logistics build is much larger than I assumed, I would not." That is exactly how professional recommendations are actually written. It also protects you: if your assumption was wrong, you have already said which one, rather than being caught defending a conclusion built on it.

What are the most common mistakes freshers make in these rounds?

Six, and they are all avoidable. Thinking silently — the interviewer cannot score reasoning they cannot hear, so narrate continuously. Jumping straight to a number without structure, which even when correct demonstrates nothing. False precision, such as 9,73,450 when every input was a guess; round numbers signal you understand your own uncertainty. Not sanity-checking, which is how candidates confidently announce that a city drinks more tea than it has people. Abandoning the structure the moment the interviewer speaks, when a question is usually a nudge rather than a rejection. And refusing to commit — trailing off with "so it depends" instead of giving a number or a recommendation with its assumptions attached. Fix those six and you are ahead of most candidates, because most preparation focuses on collecting practice questions rather than on delivery.

How do I practise this if I do not have anyone to practise with?

Alone, in three ways. First, do one guesstimate a day out loud, on your own, with paper — pick something you can see, such as the number of samosas sold in your college canteen or the number of buses on your route, and narrate the whole thing as if someone were listening. Speaking aloud is the part that must be practised; doing it in your head builds a different and less useful skill. Second, check yourself afterwards where a real figure exists — populations, vehicle registrations, airport passenger numbers are often published — not to be right, but to learn which of your assumptions was off and by how much. Third, keep a short log of your anchor numbers and the archetypes you have done, and revisit it, since four or five worked examples held firmly in your head is what makes a new question feel familiar. Then, if you can, do a few with another person interrupting you, because being interrupted is the condition you will actually perform in.

How are these asked in a group discussion versus a one-to-one interview?

The content is the same and the behaviour assessed is different. In a one-to-one, everything depends on your narration — the interviewer is inside your reasoning with you, and pace and clarity matter most. In a group setting the assessment includes how you work with others: proposing a structure the group can use is far more valuable than producing a number, and so is building on someone else's branch, keeping the group from spiralling on one assumption, and summarising where the group has got to. What fails in a group is dominating, and equally, going silent because someone spoke first. If the group is stuck arguing about whether a figure should be 50 or 60 per cent, the highest-value contribution is "either way the answer is around two million — shall we move on to the commercial segment", which shows judgement about what actually matters. Our group discussion guide covers the wider behavioural picture.

Do engineering students get asked these, or only commerce and management candidates?

Engineering students get asked them regularly, and often arrive less prepared because the question feels unrelated to their degree. Professional-services firms, consulting practices, several analytics and product roles, and some banks use estimation and short cases across all disciplines, precisely because they are testing structured reasoning rather than domain knowledge. Engineering students often do well once they start, since the arithmetic and decomposition are familiar — what trips them up is the narration, because engineering education rewards a correct final answer while this format rewards visible reasoning. If you are from a technical background, put your practice time into thinking aloud rather than into more arithmetic. Our EY, KPMG and PwC guides describe where these rounds sit in those processes.

I am in the 2027 batch. How long does it take to get decent at this?

A couple of weeks of deliberate practice gets most students from frightened to competent, which is an unusually good return compared with almost anything else in placement preparation. A workable plan: spend the first few days learning the method and the anchor numbers, then do one guesstimate a day out loud for two weeks, covering the archetypes — population-driven market sizing, supply-side capacity, headcount-driven counts, flow or throughput, and one physical estimation. Add three or four short business cases using the profit equals revenue minus cost decomposition. Then find one person to practise with who will interrupt you, because performing under interruption is the actual test. Beyond that, the highest-return background habit is reading business coverage regularly for a few months so that when a case involves an industry you have some sense of how it makes money — that is the part that cannot be crammed and it shows immediately.

I am in the 2026 batch with a professional-services interview in two weeks. What do I do?

Prioritise delivery over coverage. Learn the five-step method until it is automatic, memorise the small anchor set, and work through five guesstimates fully out loud — ideally one of each archetype, since a new question mostly feels like one you have already done. Do three short cases entering through profit equals revenue minus cost. Then rehearse with a person, because thinking aloud under interruption is the specific skill being tested and it cannot be practised silently. Two things not to do with your remaining time: do not attempt to memorise consulting frameworks, since applying one badly is worse than not using it, and do not collect fifty practice questions you never work through properly — five questions done fully and narrated beat fifty skimmed. Finally, prepare a clear answer to why professional services and why this practice, because that question decides more of these rounds than the case does.

Don't just read Guesstimates & cases questions — get asked them

Phiny's AI interviews you on exactly these topics, follows up on weak answers, and tells you what a stronger answer looks like. Text interviews are free and unlimited.

Start a free AI mock interview

How to prepare

Where these questions get asked

Related guides