Where we start
The Analyst track builds the evidence half of a market size: defensible numbers, stated as ranges, with the method on the page. Six sessions: tonight top-down from public statistics; then bottom-up from your own data; then analogy and proxy when no table measures your market; then ranges and the tornado; then share realism; then the triangulation bench, where the two methods meet. Every widget computes in your browser. Kettle & Co, a direct-to-consumer tea brand, is constructed, and so are its shares and pilot numbers; the households and the tea spend are real rows from the Statistics Bureau of Japan.
★ What you walk out with tonight. A top-down chain for Kettle & Co's premium loose-leaf tea bought online in Japan, written link by link with each link marked real or constructed and its source named, plus three notes on the table itself: the population it covers, the items you chose, and the place whose rate you used. The chain feeds the bench in Analyst session 6.
A chain of links, each one real or assumed
5 min liveTop-down sizing multiplies a population by a rate and then by shares. Population times rate is usually measured by somebody else; the shares are usually yours. Writing the chain as separate links, each with its source, is what lets a reviewer argue with one link instead of rejecting the whole number.
LiveThe four links, and where each comes from3 min▶
Population: 34,553,907 households. The 2020 Population Census counts 55,704,949 general households in Japan, of which 21,151,042 are one-person households (38.0 percent). The rest, 34,553,907, are two-or-more-person households. That subtraction is ours; both inputs are read from Table I-4-2 of the Census results outline.
Rate: 5,737 yen a year of leaf tea. The Family Income and Expenditure Survey publishes what an average two-or-more-person household spends on each item, averaged over 2023 to 2025. Green tea leaf 3,187 yen, black tea leaf 878, other tea leaf 1,672: together 5,737 yen.
Shares: premium 33 percent, online 30 percent. Neither is in any table. They are Kettle & Co planning assumptions, constructed for teaching, each with a low and a high: premium 25 to 40 percent, online 22 to 40 percent. A real team would source them from retailer panels, its own pilot or interviews, and would say which.
Multiplied through, the middle is 19.63 billion yen a year. Carry every assumption to its low together and to its high together and the range is 10.90 to 31.72. The widget below prints both, and so does the bench in Analyst session 6.
Self-studyWhy a chain beats a single "market is X" line2 min▶
A single figure can only be accepted or rejected. A chain can be repaired. When the head of BD says "30 percent online feels high", the analyst changes one link and reprints the size in a minute, and the conversation moves to the evidence for that one share. Writing the chain also exposes double counting: if the premium share was estimated from online orders, multiplying by an online share again counts the same filter twice.
Two real numbers, two guesses, one confident slide. Kettle & Co's first Japan slide said "19.6 billion yen, source: Statistics Bureau of Japan". The household count and the spend are from the Bureau; the two shares that cut the product by a factor of ten are not. The rewritten slide shows the chain with the shares marked as assumptions, and the BD lead now knows which two numbers to go and test. Constructed Kettle & Co example.
Who the table asked, and what its words mean
5 min liveA government table answers the question its designers asked, which is rarely yours. Two checks come before any multiplication: whose behaviour the table measured, and what each item name actually includes. Both change the answer more than most assumptions do.
LiveThe population: two-or-more-person households only3 min▶
The survey table used here covers two-or-more-person households. One-person households are measured as a separate survey population and are not in the rows. So the honest chain multiplies the survey rate by the 34,553,907 households the rate describes, not by all 55,704,949.
The widget lets you include one-person households "at the same spend". Do it and the top-down moves from 19.63 to 31.64 billion yen a year. That option is an assumption, and it is labelled as one: there is no row here that says how much a person living alone spends on leaf tea. If your segment is young professionals living alone, this is the link to go and source properly, not to default.
LiveThe items: read the names, then the totals2 min▶
The survey's tea group has four items: green tea leaf, black tea leaf, other tea leaf, and bottled and canned tea drinks. Bottled drinks are 8,633 of the 14,370 yen, 60 percent of the group. A loose-leaf brand that sizes "tea" from the group total starts two and a half times too high before any share is applied.
"Other tea leaf" is the survey's catch-all leaf item. It is tempting to guess what is in it; do not put the guess in the chain. Name the item as the table names it and say whether you included it.
Self-studyPublished tables do not always add up2 min▶
In the national row the four tea items add to the tea total exactly. In 15 of the 52 city rows they differ from the published total by 1 or 2 yen, because each figure is a rounded average published on its own. A validation check that demands exact equality fails on real data and teaches the team to switch checks off. Write the tolerance into the check instead: the engine behind these widgets accepts up to 2 yen and fails beyond it.
Whose spending did you assume everyone shares?
5 min liveThe table also publishes a rate for each of 52 prefectural capitals and designated cities. That is useful for choosing where to launch, and dangerous when one city's rate is multiplied by the whole country. A launch team based in the most tea-loving city in Japan will size Japan from its own street.
LiveShizuoka, Naha, and the national row3 min▶
Green tea leaf spend runs from 7,697 yen a year per household in Shizuoka, the highest city, to 1,246 yen in Naha, the lowest, against 3,187 nationally. Across all leaf items Shizuoka households spend 9,694 yen, 1.69 times the national 5,737.
Apply Shizuoka's leaf rate to every two-or-more-person household in Japan and the top-down prints 33.16 billion yen a year instead of 19.63. Apply Naha's and it prints 11.22. Nothing else in the chain moved. City rows are the right tool for "where should Kettle launch first?"; the national row is the right tool for "how big is Japan?".
Self-studyWhen a regional rate is the right one2 min▶
If the decision is a regional launch, size the region with its own rate and its own household count, and say so in the chain. The mistake is only the mix: a local rate multiplied by a national population. The same mix happens in other forms, such as sizing a country from the spend of the customers who found you first, who are rarely average.
The launch team's home town. Kettle & Co's first Japan hire lives in Shizuoka and builds the sizing from "what households around here spend". The number comes out 1.69 times too high for Japan, and nobody notices because the source line still says "official survey". The fix is one dropdown: the national row. Constructed Kettle & Co example; the survey figures are real.
Run the chain on the real rows
★ 22 min · everyone buildsThe widget below holds the national row and all 52 city rows of the survey's drinks table, plus the Census household counts. It recomputes the chain the moment you change a control. Work the five steps in order and write down what the widget prints at each one; the shares are constructed, so once you move a slider your numbers will differ from the ones quoted here.
Step 1 · Read the default chain. Category premium loose-leaf, population two-or-more-person households, place the national row, both sliders at their middles. Copy each link and its tag, real or constructed, into your notes.
Step 2 · Change the category to "Tea, all of it". The spend link jumps to the survey's whole tea group, bottled drinks included, and the premium link disappears because the frame is no longer premium. Write down the new size, then put the category back.
Step 3 · Change the population to "All households, one-person included". The households link becomes 55,704,949, and the widget says out loud that this assumes one-person households spend the same. Note the size, and note that the assumption is now in your chain.
Step 4 · Set the population back, then change the place to Shizuoka, then Naha. Read the hint line under the chain: it tells you the rate now describes one city. Then set the place back to the national row.
Step 5 · Move each slider once, then write the chain as a sentence. Drag the premium share to where you believe it sits and read the new size; do the same for the online share. Then write the chain the way it will go in the sizing brief, one clause per link, each with its source or the word "assumed".
Keep the sentence. It is the top-down half of the triangulation bench in Analyst session 6. Analyst session 2 builds the other half from Kettle & Co's own pilot, from the bottom up, and the gap between the two is where the learning is.
Before Analyst session 2
30 min- Find the official table for your own market: a household expenditure survey, a business census, an import statistic. Write its exact name, table id, year and the population it covers.
- List its items that sit near your product, and mark which ones your product actually competes in. Check whether the items add up to the published total, and by how much they miss.
- Write your chain in the step 5 format, with every link tagged real or assumed. Count the assumed links.
- For the assumed link that shrinks the size most, write the cheapest way you could source it in a week: a retailer panel, a pilot, ten interviews.
Three questions
Clear all three and Analyst session 1 gets stamped in your passport.
1 · The survey's tea spend describes two-or-more-person households. Which household count belongs in the chain?
A rate describes the population it was measured on. Including one-person households at the same spend moves the top-down from 19.63 to 31.64, and that move is an assumption, not a statistic.
2 · A loose-leaf brand sizes "tea" from the survey's tea total of 14,370 yen. What went wrong?
Read the item names before the totals. The tea group includes bottled drinks, 8,633 of the 14,370 yen.
3 · The chain uses Shizuoka's leaf spend for every household in Japan and prints 33.16 instead of 19.63. What is the error?
City rows answer "where first?". The national row answers "how big is Japan?". Mixing them inflated the size 1.69-fold without changing any assumption.
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 1 cheat sheet · pin this
You can build a top-down chain on real rows and say which links are real. Analyst session 2 builds the same market from the other end: buyers times spend per order times orders a year, from Kettle & Co's own pilot data.