Where we start
Analyst sessions 1 and 2 built Kettle & Co's Japan household market twice, from a government table and from its own pilot. Kettle's BD lead now asks the question from Business session 1 that neither chain can reach: how big is the B2B office tea market? No government table measures what offices spend on tea at item level, and Kettle has no office customers yet. Tonight is paper and pencil, no widget: three ways to borrow a size, the failure mode of each, and an office tea estimate built from two independent proxies. Kettle & Co and every office number on this page are constructed; the household tea figures are real rows from the Statistics Bureau of Japan.
★ What you walk out with tonight. A one-page office tea estimate for Kettle & Co built from two proxy chains that do not share a signal, the gap between them written down as a range, every input tagged constructed, and a list of the real data that would replace each one. Plus a sentence per method naming the one thing that would make it wrong.
A comparable market, and the dials between it and yours
5 min liveSizing by analogy takes a market you can measure, another country, an earlier year, an adjacent product, and carries its size across to the one you cannot. The comparable is never the same market, so the carry needs adjustments: for how many buyers there are, for how much each spends, for how mature the channel is. The analogy is only as good as the adjustment on the dial that drives the size.
LivePick the comparable, then list what differs3 min▶
Three kinds of comparable. Another country where the category is already measured. An earlier year of your own market, before a channel or a regulation changed. An adjacent product sold to the same buyers in the same way. Each gives you a size you can read; none gives you yours.
Then write the adjustments as separate dials. Population: scale by the number of buyers, households or offices, in each market. Spend level: scale by what each buyer spends, which differs by income, habit and price. Channel maturity: scale by how much of the category is bought the way you sell it, online, by subscription, through a distributor. Each dial is either measured or assumed, and the page says which.
The illustration from Analyst session 1 is real. Shizuoka households spend 9,694 yen a year on leaf tea against 5,737 nationally, 1.69 times as much. If Shizuoka were your comparable and Japan your target, the population dial would be the easy part and the spend dial would decide the answer. Here both ends are in the same survey, so you can check the dial. Usually you cannot, which is why it has to be argued out loud.
Self-studyThe failure mode: the comparable differs in the one thing that drives the size2 min▶
Most analogies are adjusted for population, because population is easy to look up. The adjustment that breaks them is the one nobody looked up: a tea culture that is centuries deep in one market and a novelty in the other, a channel that is mature in the comparable and barely open in yours, a price level that buys a different product. Before you trust an analogy, write one sentence naming the factor that most drives the comparable's size, and check whether your target shares it. If it does not, the analogy is a story about the comparable.
A useful habit: put a range on each dial, not a point, and carry the ranges through. Analyst session 4 does exactly that with the tornado; an analogy with three ranged dials usually turns out wider than anyone expected.
Self-studyWhen an earlier year is the comparable2 min▶
Your own market a few years ago is often the closest comparable you have, because the buyers and the culture are the same. The dial that moves is channel maturity: online buying, delivery speed and payment habits shift faster than tastes. If the comparable year predates a change in how the category is bought, adjust for the change explicitly, and say what you assumed it did.
The home market as a template. Kettle & Co's founder proposes sizing Japan as "our home market, scaled up by population". The population dial is the only one turned. The analyst adds the other two and finds that the spend dial alone could move the answer by more than the population ratio, in either direction, because tea habits differ so much between countries. The slide now shows three dials, two of them marked assumed, and the BD lead asks for evidence on spend before anything else. Constructed Kettle & Co example.
Sizing from the neighbour, and whose buyers they are
5 min liveWhen your category is not measured but the one next door is, it is tempting to start from the neighbour and take a fraction. That works only when you can say how the neighbour's buyers relate to yours. A substitute shares your buyers and competes for their spend; a complement sits beside your product in the same basket; a category that merely shares a shelf may have no buyers in common at all.
LiveBottled tea drinks next to leaf tea3 min▶
In the same survey that gives leaf tea 5,737 yen per household a year, bottled and canned tea drinks take 8,633 yen, 60 percent of the tea group. If leaf tea were unmeasured, the obvious move would be "leaf is some fraction of bottled". The fraction is the whole argument, and it depends on whether the bottle buyer is the leaf buyer. A bottle bought from a vending machine on the way to work and a tin of leaf brewed at home may be the same person in two moods, or two different people. The survey reports spend per household; it does not report who buys both.
The test to apply: would a Kettle customer buy fewer bottles if they bought more leaf (substitute), more bottles (complement), or would nothing change (different buyers)? Only the first two make the neighbour a sizing base, and each needs its own conversion.
LiveWellness drinks: the neighbour that is too wide2 min▶
Business session 1 showed the wellness drinks frame: tea plus mineral water, lactic acid drinks and sports drinks. Every item is real and in the same table. As a neighbour to size from, it fails twice: most of its spend is categories Kettle does not sell, and the buyers of sports drinks have no stated reason to be buyers of leaf tea. A size taken as "a share of wellness" carries a share nobody can source. The wider the neighbour, the more of the answer sits in that one assumed fraction.
Self-studySubstitutes, complements, and the buyers who are not yours2 min▶
A substitute neighbour is the safest base: the buyers already spend on the need you serve, and the argument is how much of that spend could switch. A complement is weaker: it says how many baskets your product could sit in, not how often it will. A shelf neighbour, same aisle, same supplier, different need, is the weakest, and it is the one most decks use because its number is biggest. Write the relationship next to the neighbour on the page, so a reviewer can disagree with it rather than with the whole size.
"Bottled tea is huge, we only need a slice." A Kettle & Co marketing draft sizes premium leaf as 10 percent of bottled tea drinks, because the bottled number is bigger and feels safer. The analyst asks one question: what do we know about bottle buyers that makes a tenth of them leaf buyers? The answer is nothing, so the line is rewritten to start from the leaf items the survey already publishes. Constructed Kettle & Co example; the survey figures are real.
Signals that move with the market, and the conversion you must assume
5 min liveA proxy is a number somebody measured for another reason that rises and falls with your market: how often people search for it, how often firms hire for it, how much of it crosses a border. A proxy is never a size. Turning it into yen needs a conversion step you assume, and each proxy leaks in its own direction before it even reaches that step.
LiveSearch volume2 min▶
What it measures. How often a phrase is typed into a search engine, usually as an estimate from a keyword tool, per month.
How it maps to a size. Searches to searchers, searchers to buyers, buyers to spend a year. Every one of those three steps is assumed, so write them as separate dashed links.
How it fails. Search counts curiosity, not purchase: a recipe, a news story or a gift idea all show up as demand. Keyword tools model the volume rather than report it, and two tools can disagree on the same phrase by a wide margin. Treat it as a relative signal, this phrase against that one, this month against last, far more than as a count of buyers.
LiveJob postings2 min▶
What it measures. How many firms advertise a role, which shows hiring intent for a function, such as office managers, facilities staff or pantry suppliers' sales reps.
How it maps to a size. Postings to firms with that function, firms to buying offices, offices to spend a year. The first link alone hides an assumption about how many postings one firm makes.
How it fails. Postings count intent to hire, not spend; they lag the market, because firms hire after demand has grown; and they cluster in big firms that post often and publicly, while small offices hire through word of mouth and never appear. A postings-based size leans towards big offices.
LiveImport statistics1 min▶
What it measures. Goods crossing the border, by weight and declared value, classified by product code.
How it maps to a size. Imported tonnes or customs value, times a markup to retail or wholesale yen, times the share that reaches your buyer. The markup and the share are assumed.
How it fails. Import figures count trade flows, so they miss everything grown and packed at home, which for tea in Japan is a serious gap. They include goods that are re-exported. And a product code is usually wider than your product: one code may hold several kinds of tea in several pack sizes, premium and bulk alike.
Self-studyWhy one proxy is never enough2 min▶
A single proxy gives a single number with a hidden conversion, and nothing to argue it against. Two proxies built from independent signals give two numbers, and the gap between them is information: it says how much the conversions disagree. This is the same lesson Analyst session 6 teaches about top-down and bottom-up. A check only tests what the two methods do not share. Two proxies that both start from the same count of offices will agree with each other and still both be wrong about the offices.
The office tea question, from Business session 1. Kettle & Co's BD lead wants the B2B office tea market. The analyst looks for a government table and finds none that measures office purchases of tea at item level: the household survey asks households, and nothing read for this work breaks out what offices buy. So the size has to be built from proxies. A search-volume chain on office tea delivery phrases would count curiosity; an import chain would count every tea crossing the border, households and offices alike. The analyst picks two signals that share as little as possible, a count of offices and a count of office workers, builds each into yen with constructed conversions, and shows BD both numbers and the gap. Constructed Kettle & Co example; no real counts are used.
Size Kettle's office tea market by proxy, on paper
★ 22 min · everyone buildsNo widget tonight: a pencil, a calculator and the five steps below. Each step gives a template with blanks and a Kettle & Co example. Every input in the example is a constructed round number chosen to make the arithmetic easy to follow, not a count of anything real; the method is what carries over. Fill the template for your own market as you go, and write "constructed" next to every input you have not sourced.
Step 1 · Define the category and the buyer, and say whose call it is. What exactly would Kettle sell to an office, to whom, instead of what? That is a category choice, and Business session 3 assigns it to BD, not the analyst. Write BD's answer at the top of the page before any number, because both proxies inherit it.
Step 2 · Proxy one: count offices. Offices in the segment, times the share that provide tea in the pantry, times what one such office would spend on leaf tea a year. All three are constructed in the example. Do the arithmetic in full, units and all.
Step 3 · Proxy two: a different signal. Start from something proxy one did not use. The example counts people and cups instead of offices and budgets. A search-volume or import chain would also do, as long as it does not quietly reuse the office count from step 2.
Step 4 · Compare the two and write the gap. Divide the larger by the smaller. Then name the input most likely to explain the gap: the one you trust least, or the one where the two chains imply different things about the same office.
Step 5 · Write which real data would replace each constructed input. One line per input. This list, not the range, is the most useful thing you hand BD: it turns "the market is 2 to 3.6 billion" into "here is what we need to find out, and roughly what it costs to find".
Keep both chains, not just the range. Analyst session 4 turns each input into a range and asks which one moves the answer most. With two proxy chains you can do that for both and see whether the same input dominates in each, which tells you whether the gap is about the market or about your conversions.
Before Analyst session 4
30 min- Pick a comparable for your own market, another country, an earlier year or an adjacent product, and write its three dials: population, spend level, channel maturity. Mark each measured or assumed, and name the one that most drives the comparable's size.
- Name your nearest measured neighbour category and write, in one sentence, whether it is a substitute, a complement or only a shelf neighbour, with your reason.
- Build two proxy chains for a market no table measures, from two signals that share as little as possible. Tag every input constructed or sourced, and write the gap between the two results.
- Write the replacement list from step 5 for your chains, and put the cheapest replacement first.
Three questions
Clear all three and Analyst session 3 gets stamped in your passport.
1 · A team sizes Japan's leaf tea market from Shizuoka's household spend, adjusting carefully for the number of households. What is the failure?
An analogy needs every dial that drives the size. Here the spend level dial is 9,694 against 5,737 yen; population was the easy dial and the wrong one to stop at.
2 · A deck sizes premium leaf tea as "10 percent of bottled tea drinks, a much bigger category". What should the reviewer ask first?
A neighbour is a base only if its buyers relate to yours. The survey reports spend per household, not who buys both, so the fraction is pure assumption until that relationship is argued.
3 · An office tea size rests on one proxy: search volume for "office tea delivery" times assumed conversion rates. What is the best next move?
Every proxy leaks. A second keyword tool shares the same leak; import data has its own and misses domestic production. Two independent chains give a range and show which conversion to source.
What this session teaches, and where it came from
Every fact on this page, and its verification tier, is recorded in the course's source map.
Analyst session 3 cheat sheet · pin this
You can size a market no table measures and say how each borrowed number could be wrong. Analyst session 4 makes that precise: every input becomes a range, the corners give the size a low and a high, and the tornado shows which input moves the answer most.
Whether Kettle can actually win the office tea it just sized is a pipeline question, not a sizing one: B2B Sales Pipeline Analytics ↗ picks up exactly there.