Market sizing: estimating the unknowable

Intuition

Market sizing feels impossible, "how many tennis balls are sold in the UK each year?", until you realize nobody expects you to know. They expect you to build. It's like estimating how many jellybeans are in a jar: you don't count them, you figure out the volume of the jar, the volume of a bean, and divide. The magic is the method, not the memory.

A good estimate is just a chain of reasonable assumptions, each one said out loud so the interviewer can nod along.

Framework

• Choose top-down or bottom-up. Top-down starts from a big population and filters down. Bottom-up builds from a unit (one store, one customer) and scales up. Pick whichever has a cleaner anchor.
• Segment when it changes the rate. Split the population only where it affects the number (e.g. coffee drinkers vs non-drinkers).
• Use round, defensible numbers. US population ≈ 330M. Round so the math stays clean, precision is fake comfort here.
• Sanity-check the answer. Convert it to a per-person or per-day figure and ask "does that feel real?"

Worked Example

"How many smartphones are sold in the US per year?" Top-down: ~330M people → ~80% own a smartphone → ~265M phones in use → people replace roughly every 3 years → ~265M / 3 ≈ 88M phones per year. Sanity check: that's about one phone per four people per year, which feels right for a mature market. Every step is a number the interviewer can challenge, and you can defend.

Now flip to bottom-up: "What's Starbucks' US retail revenue?" One store serves ~600 customers a day at ~$7 a ticket ≈ $4, 200/day, call it $1.5M a year. Starbucks runs ~16, 000 US stores, so 16, 000 × $1.5M ≈ $24B. Same skill, opposite direction: top-down filters a population down; bottom-up scales up a unit you can picture. Pick whichever anchor you trust more, for a single chain, the store-level unit is far firmer ground than a population filter.

In the room

Buy structure time before any numbers: "Let me take 30 seconds to lay out my approach before I calculate." Then walk the path skeleton first, population, filter, rate, and only then compute. Flag every assumption as a checkpoint: "I'll assume 80% smartphone penetration, happy to adjust if you have a better figure." That sentence invites correction without surrendering control. When the arithmetic comes, narrate the operation before the result: "330 million times 80 percent… is about 265 million." Saying the operation first means a pause reads as computing, not panicking. Land it with the sanity check spoken aloud: "That's roughly one phone per four people per year, feels right for a mature market." At that moment the interviewer is scoring one thing: whether your number came from a defensible chain of logic they could audit, not from luck.

Pitfalls

• Diving into arithmetic before laying out the path.
• Over-segmenting (splitting by age when age doesn't change the answer) and drowning in steps.
• Reporting a wild number without the sanity check, order-of-magnitude errors are the most common fail.