When a sizing prompt drops, I want a flow I can run fast and defend. My approach lately: lock the scope in 15–20 seconds, pick a clean top-down anchor, then cross-check with a bottom-up driver that hits the real unit economics. If the two land within the same order of magnitude, I keep moving. If not, I run a quick sanity pass using per-capita spend bounds or an analog company’s revenue to avoid clown numbers. I’m not trying to be clever—just crisp and believable under the clock. What’s your exact sequence under 10 minutes, and which sanity checks have saved you most often?
stop worshipping frameworks and start using math that won’t embarrass you. pick a top-down base rate you can say out loud without blushing, then build a bottom-up using the single driver that actually matters (units or frequency, not 8 things). if they’re a mile apart, your assumption’s junk. sanity check with per-household spend and a public comp’s revenue. if your number implies $1k/yr on dish soap per person, congrats, it’s wrong. leave 60 seconds to round and narrate. no one cares about your mece art, they care it’s defensible.
the fastest trap: stacking tiny errors. pick a constraint and hammer it. examples: installed base x replacement cycle, or % of disposable income that could plausibly go to the category. then compare to a real company’s last 10-k. if your market is 3x p&g’s fabric care revenue in france, lol, no. also, define the buyer once. flipping between users and purchasers mid-case is how you invent imaginary dollars. keep it boring, keep it tight.
quick take
i start with “who buys + what unit,” top-down anchor, then 1 bottom-up driver (units x price x freq). 7‑min timer. 2 sanity pings: per-capita spend + public comp. i still wobble narrating the switch. tips to make that smoother?
practice ask
anyone up for 10‑min mocks tonight? i’ll host a zoom. 2 prompts, swap roles, fast feedback. promise to timebox and catch 10x errors. dm if in!
Two moves keep this tight and credible. First, state a single spine before you compute: “I’ll anchor top-down on adult population, then validate bottom-up using units per buyer and price.” This pre-commits you and prevents detours. Second, define a hard boundary up front: either per-capita spend ceilings or a share-of-wallet for the category. After your first pass, triangulate: if top-down is 25–30B and bottom-up is 18–22B, you’re close enough; reconcile the big delta with one explicit assumption (e.g., penetration or replacement cycle). Close by verbalizing the risks and upside cases in one sentence so it sounds intentional, not lucky.
I bombed a retail sizing because I chased five drivers and still landed 10x off. The fix came in a fintech case where I forced myself to do one top-down and one bottom-up, then ran a simple budget sanity check: could a typical user even afford that annual spend? That caught the error instantly. Interviewer literally said, “Thanks for calling your own bluff.” Since then, I keep a go-to analog revenue in mind and a per-household cap. It’s not fancy, but it travels across industries.
A compact, defensible sequence: 1) Scope precisely (buyer, geography, year). 2) Top-down: start with a known base rate (population, firms, or installed base) and apply penetration and spend per unit. 3) Bottom-up: core unit count × price × replacement frequency. 4) Sanity checks: compare against a public comp’s segment revenue and a realistic share-of-wallet or per-capita ceiling. 5) Reconcile variance by adjusting one named assumption. Narrate each assumption with a quick analog (e.g., “replacement similar to smartphones, ~3 years”) to keep it grounded.