CASE INTERVIEW
“How many cups of coffee are sold in Toronto each day?”
You know absolutely nothing about the Toronto coffee market.
No dataset.
No paper.
No Google.
For a scientist, that can feel absurd.
Then you realize you have been doing versions of this for years.
Market sizing is not really about knowing the answer.
It is about building one from what you can reasonably estimate.
The question sounds commercial... the logic should feel strangely familiar
Suppose you need to estimate annual coffee-shop purchases in a city.
You could panic because you do not know the number.
Or you could decompose it.
Population.
Share who drink coffee.
Share who buy from cafés.
Purchases per week.
Weeks per year.
Multiply.
That structure is not far from estimating cell yield, reagent consumption, sample throughput, animal numbers, sequencing capacity, or experimental cost.
You identify the variables.
You estimate inputs.
You calculate the output.
The unfamiliar part is the subject, not the reasoning.
If you want the mechanics in detail, read the full Market Sizing guide.
But your scientific precision can suddenly become the trap
Here is where the comfortable analogy breaks.
In research, you may spend substantial effort improving an uncertain input.
Is it 63% or 68%?
Can we find a better source?
Should we report confidence intervals?
In a case interview, the interviewer may not care whether your coffee-drinking estimate is 60% or 65%.
They care whether the assumption is reasonable, explicit, and easy to work with.
False precision wastes time.
If the population is roughly 3 million, use 3 million unless greater precision changes the decision.
If half the population drinks coffee, say why that seems reasonable and move.
The goal is not to manufacture accuracy from missing information.
It is to create a transparent model.
There are two numbers in your calculation... and they are not the same
This distinction catches strong quantitative candidates surprisingly often.
Imagine you say:
“Toronto has 3 million people.”
That may be an estimate based on general knowledge.
Then:
“I will assume 50% buy coffee outside the home.”
That is an assumption you introduced to make the model work.
Those are different.
Why care?
Because when the final number looks strange, you need to know what to challenge.
Was your estimate of population wrong?
Or was the behavioural assumption too aggressive?
For a fuller treatment of this distinction, read Assumptions vs Estimates in Case Interviews: Why Mixing Them Up Hurts Your Logic.
The arithmetic is rarely the most interesting part
You calculate:
3 million people
× 50% café buyers
× 2 purchases per week
× 50 weeks
That gives 150 million purchases annually.
Great.
Now what?
This is where many candidates stop.
A number without an implication is unfinished work.
Suppose the client needs 15 million annual purchases to hit its target.
Your estimate implies the client needs roughly 10% market share.
That is the consulting turn.
Suddenly the market size connects to a decision.
You do not need the “right” answer as much as you think
Two candidates can make different assumptions and reach different market sizes.
Both can still demonstrate strong reasoning.
What matters is that the structure is logical, assumptions are sensible, math is controlled, and the result passes a sanity check.
That last part should feel familiar.
Scientists do not blindly trust an output because the equation returned a number.
Neither should consultants.
If your calculation says every resident buys twelve coffees a day, something broke.
Check the units.
Check the assumptions.
Check the order of magnitude.
Then move.
You already know how to model uncertainty.
The interview is simply asking you to do it without a methods section.
And with a client waiting for the answer.