Guess- timates.

Key terms

Estimation question

An interview question that asks for a number you can’t look up, like a market size or a count. Also called a guesstimate, a market-sizing question, or a Fermi problem.

Fermi problem

An estimate built from rough, reasoned guesses about its parts. For a very large number, landing within about a factor of ten of the true value is usually good enough. Named after the physicist Enrico Fermi.

Top-down

Starting from a big number you know, like a country’s population, and narrowing it step by step to the thing you want.

Bottom-up

Starting from one small unit, like one store’s sales in a day, and multiplying it up.

Assumption

A value you choose and say out loud, with a short reason, because you can’t look it up.

Sanity check

A quick test of whether your answer is believable: a second approach, or a comparison with something you know.

Order of magnitude

Roughly how big a number is, in powers of ten: is it about 1 million, 10 million, or 100 million? A million and ten million are one order of magnitude apart.

The idea

“How many sofas are bought in the US each year?” Nobody expects you to know. That’s the point. The question tests whether you can take something that looks impossible, break it into pieces you can reason about, and get to an answer that’s roughly right, out loud and in a few minutes. It isn’t a trick question, either: there’s no clever answer to spot, only a path you can show.

Product managers do this for real, too. Before a team spends months on a feature, someone has to ask: how many customers could this help? This lesson teaches a seven-step method, with one example worked all the way through: sofas bought in the US, then sofas the made-up furniture store might sell in one city.

The lesson, in seven parts

1. 1What an estimation question is, and why it’s asked.

   Estimation questions are older than product management. In physics teaching they’re called Fermi problems, after Enrico Fermi, who, in the words of MIT physicist Peter Dourmashkin’s textbook, “was famous for making these sorts of ‘back of the envelope’ calculations.” For a very large number, the same textbook says, an estimate within an order of magnitude “should be satisfactory.”

   A famous example is Fermi’s own. At the first atomic test, in July 1945, he dropped small pieces of paper as the blast wave passed and watched how far they moved. In his report he wrote that the shift was “about 2½ meters, which, at the time, I estimated to correspond to the blast that would be produced by ten thousand tons of T.N.T.” The official figure, set in 1963, is 21 thousand tons, and a 2021 Los Alamos reanalysis put it nearer 25 thousand. He was off by a factor of two to two and a half, from a handful of paper and some reasoning. That’s the skill.

   In an interview, the number matters less than the path to it. Interviewers usually want to see that you:

   • Pin down what’s being asked before you start.
   • Break a big unknown into parts you can reason about.
   • Say your assumptions out loud, with reasons, so anyone can challenge them.
   • Handle the math calmly, in round numbers.
   • Check your own answer, instead of waiting to be told.

2. 2The method, in seven steps.

   Here’s a method you can use on any estimation question. Learn the order, then practice it until it’s automatic:

   • Clarify: what exactly are we counting, where, and over what time?
   • Pick a direction: top-down from a big known number, or bottom-up from one unit.
   • Write the formula in words, before any numbers. Check it with the interviewer.
   • State an assumption for each part, with a short reason.
   • Do the math in round numbers, out loud.
   • Sanity-check: a second approach, or a comparison with something you know.
   • Summarize: one number, a range, and the assumption that matters most.

   Writing the formula is the step that’s easiest to skip under pressure, and it matters most. Saying “I’ll estimate households, times the share that own a sofa, divided by how many years a sofa lasts” lets the interviewer redirect you before you’ve spent five minutes on the wrong path.

3. 3Clarify what you’re estimating.

   “How many sofas are bought in the US each year?” sounds clear until you try to answer it. Does a loveseat count? A hotel buying 200 sofas? A secondhand sofa? Each choice changes the number, so ask, or say what you’ll assume. Two or three questions are plenty.

   Ask

   • “Should I count new sofas only, or secondhand ones too?”
   • “Just households, or businesses like hotels and offices as well?”
   • “Is a loveseat a sofa for this?”
   • “Is this the number of sofas, or the number of people or households buying one?”

   Avoid

   • Ten questions in a row. It looks like stalling.
   • “Do you know the population of the US?” Bring your own starting number instead, even a rough one, and say where it comes from.
   • Asking nothing, then estimating something slightly different from what they meant.

   For the worked example, we’ll agree: new sofas only (loveseats count), bought by households, in one year, in the US. And we’ll count purchases, assuming one sofa per purchase. And to keep it simple, we’ll count one main sofa per household, and say so out loud.

4. 4Choose top-down or bottom-up.

   There are two directions to come at almost any estimate. Top-down starts from a big number you know and narrows it. Bottom-up starts from one small unit and multiplies it out:

   	Top-down	Bottom-up
   Starts from	A big known number: population, households	One unit: one store, one customer, one day
   Sofa example	Households × share with a sofa ÷ years each sofa lasts	Sofas a store sells a day × days open × number of stores
   Good when	You know the big number well	You can picture the small unit clearly
   Watch out for	Too few steps, so one wild guess swings everything	Not knowing how many units there are

   Neither is “correct”. A good habit is to answer with one and check with the other.

   For sofas, top-down is easier: you can know roughly how many US households there are, but you probably don’t know how many furniture stores there are. So we’ll answer top-down, and check with a second approach.

5. 5Write the formula, then state your assumptions.

   Say the formula in words first: sofas bought per year = households × share of households with a sofa ÷ years a sofa lasts before it’s replaced. The logic: if most households have a sofa, and each one replaces it every so many years, then each year roughly one in that many households buys one.

   Now give each part a value, and say which ones are facts and which are your assumptions:

   Part	Value	Fact or assumption	Why it’s reasonable
   US households	About 135 million	Fact: 134.8 million in 2025 (US Census Bureau, Current Population Survey)	A round number worth remembering for US questions
   Share with a sofa	90%	Assumption	Most homes have one; some single rooms and student homes don’t
   Years a sofa lasts	10	Assumption	Many people keep a sofa for years, and some replace it sooner when they move

   The household figure is a real Census number. The other two values are assumptions made up for this example, and you should say so out loud.

   The template

   I’ll estimate [what] as [part A] × [part B] ÷ [part C]. For [part A], I’ll use [value], because [reason]. For [part B], I’ll assume [value], because [reason]. For [part C], I’ll assume [value], because [reason]. That gives [rounded math], so about [answer].

   Filled in, for the made-up furniture store

   I’ll estimate sofas bought per year as households × the share with a sofa ÷ years a sofa lasts. For households, I’ll use about 135 million, from the Census. For the share, I’ll assume 90%, because most homes have one. For the life of a sofa, I’ll assume 10 years, because people keep one for years but replace it sooner when they move. That’s 135 × 0.9, about 120 million, divided by 10, so about 12 million sofas a year.

   Then the math, in round numbers: 135 million × 0.9 is about 120 million households with a sofa. Divided by 10 years, that’s about 12 million sofas a year. Say each step as you go, and keep the zeros straight by thinking in millions.

6. 6Sanity-check it a second way.

   Before you say you’re done, check. Two quick checks work well.

   First, check your starting number against another fact. The Census Bureau estimated the US population at about 342 million on July 1, 2025, and its survey puts the average household at 2.5 people in 2025. 342 million ÷ 2.5 is about 137 million households, close to the 135 million we used. Good.

   Second, estimate the answer a different way: by the moments that make people buy a sofa. Moving home is a big one. The Census Bureau’s survey counted about 30 million movers out of about 322 million people aged one and over in 2019 to 2020, roughly 9%. Then:

   Who buys	Calculation	Sofas a year
   Households that move	135 million × 9% (assume households move at the same rate as people) × 1 in 4 buy a sofa (assumption: many movers bring the sofa they have)	About 3 million
   Households that don’t move, replacing a worn sofa	About 123 million × 1 in 15 each year (assumption: people who stay put tend to keep a sofa longer than 10 years)	About 8 million
   Total		About 11 million

   The mover rate is a real Census figure for 2019 to 2020; the rest are assumptions for this example.

   Eleven million from one direction, twelve million from the other. They don’t prove the answer is right, since both rest on assumptions, but two different paths landing close together is a good sign. As a rough guide, within two or three times of each other is close. If they’d been ten times apart, you’d look for the assumption that’s off, and say so out loud.

7. 7Summarize, and bring it down to one store.

   End with a clear, short summary: the number, a range, and the assumption that matters most. Something like: “About 12 million sofas a year, probably between 10 and 15 million. The biggest uncertainty is how often people replace a sofa: if it’s every 8 years instead of 10, the answer goes up by about a quarter.” To get the range, try a low and a high value for your shakiest assumption. A sofa that lasts 12 years gives about 10 million; one that lasts 8 gives about 15 million.

   A follow-up often brings it closer to the business. Say the interviewer asks how many sofas the made-up furniture store might sell in one city a year. Reuse your formula, and add one step:

   Step	Value	Fact or assumption
   People in the city	1 million	Assumption: a made-up city
   Households	1 million ÷ 2.5 = 400,000	2.5 people per household is the 2025 US Census average
   Sofas bought in the city each year	400,000 × 0.9 ÷ 10 ≈ 36,000	Same assumptions as before
   The store’s share	5% ≈ 1,800 sofas a year	Assumption: made-up market share
   Per week	1,800 ÷ 52 ≈ 35	Arithmetic

   A made-up example. The store, the city, and its market share are invented for this lesson.

   About 35 sofas a week. That’s a number a product manager can use: it tells you how many deliveries a week to plan for, and how many shoppers a new doorway-size check could help.

With AI: AI can look up a population in seconds and run any calculation you like, so you might wonder why this skill still matters. It matters because the hard part was never the arithmetic. It’s choosing a sensible way to break the problem down, and knowing which assumption to question. AI tools can also produce confident numbers that are wrong. The habit this lesson teaches, checking a figure a second way and asking “does that seem right?”, is exactly how you catch them.

QUESTION:

1. CLARIFY
   What exactly are we counting?
   Where?            Over what time?
   What's in, what's out:

2. DIRECTION
   [ ] Top-down (start from a big known number)
   [ ] Bottom-up (start from one unit)

3. FORMULA, IN WORDS
   ______ = ______ x ______ / ______

4. ASSUMPTIONS
   Part | Value | Fact (source, year) or assumption | Why

5. MATH (round numbers, units on everything)

6. SANITY CHECK
   Second approach:
   Compare with something I know:
   Do they land close together (within about 2 or 3 times)? If they're 10 times apart, which assumption is off?

7. SUMMARY
   About ______ , probably between ______ and ______.
   The assumption that matters most: ______
   If it changes to ______, the answer becomes ______.

Ask AI about this lesson

All lessons