How would you size a national coffee chain's annual revenue both top-down and bottom-up?

## Two directions into the same number Every back-of-envelope estimate can be attacked from either end, and a strong answer uses both. **Bottom-up** starts from the smallest repeatable unit of the business and scales it up. For a retail chain that unit is one store on one day. You estimate what happens in that store, then multiply by stores and by days. **Top-down** starts from a large aggregate you already have a feel for — a population, a total market, a budget — and cuts it down with shares and rates until only the target is left. They are not redundant. They rest on almost entirely different assumptions, so agreement between them is genuine evidence, and disagreement localises a mistake. ## The bottom-up chain 1. Stores: about 15,000 nationally. 2. Cups sold per store per day: a busy store serves through a long day with a heavy morning peak. Call it 500. 3. Average ticket: about 4 dollars per cup. 15,000 x 500 = 7.5 million cups a day. At 4 dollars that is **30 million dollars a day**. Multiplied by 365 days: **about 11 billion dollars a year**. Notice what makes this chain trustworthy: every factor is something you could in principle observe by standing in one store for an hour. That is the signature of a good bottom-up decomposition — the inputs are *witnessable*. ## The top-down chain 1. Adult population: about 250 million. 2. Share who buy coffee outside the home with any regularity: call it 40 percent, so 100 million buyers. 3. Purchase rate: about 2 cups a week each. 4. 100 million x 2 = 200 million cups a week; x 52 weeks = **about 10 billion cups a year** nationally. 5. At 4 dollars a cup, the out-of-home coffee market is roughly **40 billion dollars a year**. 6. The chain's share of that market: call it 25 percent, giving **about 10 billion dollars**. This chain rests on *per-capita anchor rates* — cups per person per week, share of the population that participates. Those anchors are the top-down equivalent of witnessable facts: you can check them against your own behaviour and the behaviour of people you know. ## Reconciling 11 billion versus 10 billion is agreement. In an interview that is the moment to say so explicitly: two independent chains converging is much stronger than one chain stated confidently. The more instructive case is disagreement. Suppose bottom-up had given 11 billion and top-down 2 billion. The gap is a 5x factor and it has to live in a specific place. The productive move is to ask which single factor, changed to a plausible alternative, closes it: - **Market share** is a classic culprit; assuming 5 percent instead of 25 percent moves the top-down answer by exactly 5x. - **Cups per store per day** is the classic culprit on the other side; 500 is a busy-store figure and the average store across a large chain is quieter. - **Participation rate** — what fraction of adults really buy coffee out — is easy to set far too low or too high from your own habits. When you find the factor, you have learned something the single estimate could never have told you: which assumption the answer actually hangs on. ## Which direction to lead with - Lead **bottom-up** when the unit is concrete and countable and the aggregate is fuzzy — a store, a truck, a technician, a machine. - Lead **top-down** when a credible aggregate exists and the unit is heterogeneous — total advertising spend, total population, a national budget. - When the two disagree wildly and you can only trust one, prefer the chain whose factors are closest to things you have personally observed. Guessing a national market size is guessing; guessing how many people queue in a coffee shop at 8am is recall. ## Common failure modes **Running the same chain twice.** If your top-down step is *market size = stores x cups x price*, you have not done a second estimate, you have relabelled the first. The two chains must not share their dominant factors, or agreement proves nothing. **Treating agreement as confirmation when a factor is shared.** Both chains above use 4 dollars a cup. That shared factor cancels out of the comparison — if the price is wrong, both answers are wrong together. Say this out loud; it tells the interviewer you understand what the cross-check does and does not verify. **Using peak as average.** A store at its 8am rate all day, or a flagship store as the typical store, inflates bottom-up estimates severely. Always ask whether the unit figure is the busy case or the mean case. **Stopping at the first answer.** The reconciliation is the part of this question being graded. A candidate who produces one chain and stops has answered half the question. ## The habit worth carrying Whenever an estimate matters, build it twice from different directions and compare. If the two agree, your confidence is earned rather than asserted. If they diverge, spend your remaining time on the factor that explains the divergence rather than on polishing either chain.