Top-down sizing starts from a large, more-established aggregate and narrows it with a chain of defensible ratios until it lands on the number you want. Most of the work is choosing the aggregate. It has to be (a) easier to anchor than the target number itself, and (b) connected to the target by ratios you can defend instead of invent. A good top-down chain reads like a sentence: "total organized consumer-durables retail, times AC's share of that basket, gives the AC category" (Worked Example 1).
How do you make a top-down build defensible?
Two practices separate a strong top-down build from one that merely sounds plausible. First, each filter ratio needs its own one-line defense. "AC is about 6.5% of durables spend" won't do. Try "AC is a meaningful but minority durables category behind TVs and general appliances, so a mid-single-digit-percent share of the durables basket is a defensible order of magnitude." The claim isn't that the ratio is precise, only that its order of magnitude holds up, and that's all a guesstimate ever needs. Second, carry the aggregate's own uncertainty forward. Most macro-level industry sizes are widely cited approximations. If yours is one, your final top-down figure inherits that uncertainty and shouldn't be quoted to more than two significant figures.
Top-down usually fails in one way: you pick an aggregate that isn't actually easier to know than the target. Sizing a niche B2B software category by starting from total global GDP and applying six multiplicative filters is technically top-down and practically useless. Each filter compounds the uncertainty instead of resolving it. So stop at the smallest aggregate you can really defend a share of. That's usually one or two filters away from a category you already have a real anchor for. A chain of five specialized guesses stacked on a single macro number is what you're trying to avoid.
Advertisement
Bottom-up sizing: the atomic-unit method
Bottom-up sizing starts at the other end. Pick the smallest unit that actually transacts (a person, a household, a store, an order, a policy, an installation) and build the aggregate by multiplying that unit's behavior by however many units there are. In Worked Example 1, the bottom-up build starts from urban households and their AC-buying behavior. Worked Example 3 starts from test-takers and their coaching spend. Worked Example 4 starts from wearable-owning adults and, separately, from retail doors and their monthly sell-through. So bottom-up is a family of unit-level builds, and picking the right atomic unit for the question is half the skill.
Bottom-up is usually seen as the more defensible of the two core methods, because every assumption inside it is concrete and can be checked on its own. How many urban households there are, what share own an AC, and how often an AC gets replaced are each things a real person could reason about from lived experience or a quick benchmark. A single opaque penetration ratio applied to an enormous aggregate gives you nothing like that. The cost is more steps. A bottom-up build is typically a chain of three to five multiplied factors, and each extra factor widens the band of plausible error. Building a range puts numbers on that.
To choose the atomic unit well, ask what's the smallest thing that, multiplied by "how many of them are there" and "how they behave," rebuilds the whole market with no double-counting and no gap. For a durable good bought once per household, that's a household. For a repeat-purchase service, it's a transacting user. For a subscription business it's a policy or a cohort, the unit the unit-economics lens already trained you to reach for. For a B2B per-seat product, it's employee headcount. Get the unit wrong, say by counting people for a product bought once per household, and the bottom-up build quietly overstates the market by the average household size. That mistake looks like careful arithmetic, but the error is in the scoping, one level up.
Two more routes: supply-side capacity, and analogy/comparables
Top-down and bottom-up cover most guesstimates. Two more techniques earn their place often enough to carry as named tools, and each one solves a problem the first two can't.
Supply-side / capacity sizing asks a different question altogether. Set aside how much buyers want, and ask how much the industry's actual delivery capacity could support. You build it from the producing or serving side: the number of factories, crews, professionals, or machines that can deliver the product or service, times their realistic throughput per year. Worked Example 2 uses installer workforce (crews) times installations per crew per year as a third, independent cross-check on a residential-solar demand estimate. Practice P12 sizes an audit-services market from the realistic engagement capacity of the chartered-accountant workforce. Capacity sizing earns the most where demand-side assumptions are softest. Think of a nascent or supply-constrained category, where "how many buyers want this" can't be known but "how many could physically be delivered this year" is bounded by countable inputs like headcount, crew size, working days, a realistic job-per-crew rate. It also works as a natural sanity check on any demand-side estimate that implies more volume than the industry could conceivably produce. Say your top-down and bottom-up both come to 5 million installations, and the entire installer workforce couldn't physically complete 500,000. Then one of your demand-side numbers is wrong, however carefully it was built.
When should you size a market by analogy?
Analogy / comparables sizing solves the opposite problem: a category so new there's no established base rate to anchor a penetration assumption on. Don't invent a number. Borrow a plausible adoption pattern from a comparable earlier category, such as an adjacent wearable device, an earlier wave of similar technology, or a similar product's early-year penetration among the same enthusiast population, and apply that borrowed ratio to the new category's addressable base. Worked Example 4 sizes a nascent wearable-device category this way. It borrows the order-of-magnitude early-adoption penetration that a comparable earlier wearable category reached at a similar stage, and applies it to adults who already own some fitness wearable, the natural first buyers for a new one. Practice P15 makes the same move for a still-newer category, and deliberately uses a lower borrowed penetration rate because the analogy is one step further removed.
Analogy sizing is the softest of the four methods. Here you assert the category will rhyme with a different one instead of measuring it, and that assertion can simply be wrong, since a truly novel product can adopt faster or slower than any precedent. Two practices keep it from being reckless. First, name the analogy explicitly ("borrowing early-adoption penetration from category Y, not asserting this is category Y"), so a listener can argue with the comparison itself as well as the arithmetic downstream of it. Second, widen your stated range more aggressively than you would for a demand-funnel bottom-up. A well-anchored bottom-up might defend a ±20–30% band. An analogy-based estimate should carry something closer to ±50–100%, because the largest single source of error (is the analogy even valid?) sits outside the arithmetic entirely. Reach for analogy sizing only when top-down and bottom-up both fail for lack of any anchor. Of the four methods, it's the last resort.