Estimation and Market-Sizing Questions

Problem This Solves

Candidates panic when asked open-ended quantitative questions ("How many pizzas are sold in Manhattan each year?") and either refuse to engage, apologize for math skills, or jump straight to arithmetic without a structure. Interviewers are not testing factual recall; they are evaluating whether the candidate can decompose an ambiguous problem, state assumptions, and reason transparently under uncertainty — all core PM skills.

Key Principle

Process over answer. "Estimation questions are entirely about the process you take to solve them. It's the journey, not the destination." Being off by a factor of two from the true census figure does not disqualify you. Stating an assumption without justification, or skipping the sanity check, does. Your reasoning is more important than the precise number.

The 8-Step Framework

1. Clarify the question — Repeat it back; resolve ambiguity. Profit or revenue? US only or worldwide? Shampoo only or conditioner too? Manufacturer revenue or reseller revenue?
2. Catalog what you know (or wish you knew) — List known facts; selectively ask the interviewer ("Could you tell me X, or would you prefer I estimate it?").
3. Make an equation — Express the answer as a product/quotient of estimable quantities before touching numbers. Example: [Gmail US users] x [annual clicks per user] x [avg revenue per click].
4. Think about edge cases and alternate sources — Identify what the main equation misses (illegal gun sales, college towns, freight elevators). Name them proactively; your interviewer has asked this question dozens of times and knows the gaps.
5. Break it down — Solve each equation component with its own sub-equation. Keep steps separate so errors are isolatable.
6. State your assumptions — Commit to each number aloud with a rationale. "Shampoo runs about $5/bottle for women. Stores mark up ~50%, so we'll work with that."
7. Do the math — Use round numbers. Shoot for the right order of magnitude, not false precision.
8. Sanity check — After computing, validate the result with an alternative reference point. If Gmail ad revenue comes to $16 per US person, something is wrong. Catch your own errors without prompting.

Good Examples

Shampoo industry revenue (full worked example)

• Clarified: revenue (not profit), shampoo only (not conditioner), manufacturer revenue (not reseller margin).
• Equation: [US population] x [% who use shampoo] x [bottles/year] x [price/bottle at manufacturer level].
• Segmented by gender; used 1 tsp per shower, ~100 showers per bottle, $5/bottle women, ~$3 men, 50% store markup.
• Computed ~$2B. Sanity check: 3.5 bottles/year = 3.5 oz/month; a travel-size bottle (1 oz) lasts ~2 weeks, so usage is under-counted by ~2x. Revised upward, then adjusted back down for bald individuals and infrequent washers. Final estimate: ~$2B.

Dog food annual spend

• Segmented households into "with children" (30% of households, 30% dog ownership rate) and "without children" (70% of households, 10% dog ownership rate) to produce a blended 16% dog-ownership estimate.
• Actual rate was 39% — the segmentation approach correctly isolated the key driver of error (families with children own dogs at a much higher rate).

Tennis balls in a two-bedroom apartment

• Pitfall avoided: used effective volume (diameter x diameter x adjusted height = 2.5 x 2.5 x 2 = 12 in³ = 1/144 ft³) rather than sphere volume. "Many people approach this by using the volume of a sphere. This is not quite correct."

Bad Examples

• Self-deprecating about math ("I'm not great at numbers, but..."): explicitly counterproductive. Never do this.
• Jumping to arithmetic before forming an equation: produces a single large computation that is impossible to debug when the sanity check fails.
• Using a single average for a heterogeneous population: "How many dogs are in the US?" with a flat ownership rate misses the dominant driver (family composition). Segment first.
• Skipping the sanity check: the police officer estimate produced 1.2 million officers; the candidate noted 1-in-150 men being police seemed high and flagged it. Interviewers reward self-correction.

Key Quotes

"I promise you: no one actually cares if you know how many pizzas are sold every year in Manhattan. Even if you knew the 'right' answer, it wouldn't help you. In fact, it could distract you."

"Estimation questions are entirely about the process you take to solve them. It's the journey, not the destination."

"Your reasoning is more important than the precise number."

"Remember that it's the approach that matters, not the final answer."

Rules of Thumb

Standard reference numbers (memorize these):

• US population: 300 million
• US households: 100 million (avg 3 people/household)
• US life expectancy: 80 years; World: 65-70 years
• World population: 7 billion; Europe: 700 million; Asia: 4 billion
• Hours in a year: ~9,000; Minutes in a year: ~500,000

Math shortcuts:

• Use round numbers throughout (314 million -> 300 million); simplifies arithmetic without sacrificing order-of-magnitude accuracy.
• Rule of 72: years to double = 72 / annual growth rate %. At 9% salary growth: 72/9 = 8 years (precise: 8.05 years). Use 70 or 75 as substitutes for mental division.
• Order-of-magnitude check: digits(a) + digits(b) should equal digits(a x b) or digits(a x b) + 1. If off by more, a zero-counting error occurred.
• Label all units on every value; unit-confusion bugs are "very difficult to find."
• Keep each computation step on a separate line so errors can be isolated without restarting.
• Before the interview, look up the target company's revenue, profit, and user count — ready-made anchors for estimation problems.

Benchmarks from worked examples:

Related References

• Product Design and Improvement Questions - Product questions that follow estimation
• Case and Strategy Questions - Case questions that use similar decomposition