Case studies
A Japanese pre-owned watch retailer, measured — and how the same shape applies in other industries.
Measured: a pre-owned watch retailer
The company is not named. It is a Japanese retailer of pre-owned luxury watches. Every figure below was measured; none was invented for illustration.
Where they started
They knew their own inventory. Cost, listing price, days on hand — all of it sat in their own systems. What did not sit anywhere was where that inventory stood in the world this morning.
They knew foreign buyers were purchasing through parallel channels, and that walk-in customers at the Japanese stores were often visitors. So a price gap existed. Which references carried the gap, and how large it was, is published by nobody.
What was measured
Twelve model families, read every morning across 50 sources in Japan and overseas. Every row records the seller, the listing URL, the moment it was read, and the exchange rate applied, along with the date that rate was published.
| Measurement | Value |
|---|---|
| References matched across markets | 44 |
| Inventory reconciled (ex-tax) | 777 items / ¥2,036,585,182 |
| Japan → United States, median gap | +9.2% |
| Japan → Hong Kong, after tax normalisation | −4.3%, with 4 references *dearer* in Hong Kong |
| Export candidates in the high-confidence band | 156 items / ¥55,749,833 |
Why these numbers are not ordinarily available
Mixing tax-inclusive and tax-exclusive flips the sign
The first pass showed Hong Kong 12.9% cheaper. The real gap is 4.3%. Hong Kong levies no consumption tax; the Japanese listing price includes 10%. Two different things, both called "price", were being subtracted.
After the correction, 4 of 16 references turned out to be dearer in Hong Kong. Acting on the first number would have meant exporting those four into a worse market.
The same error ran the other way in the United States, making the real gap look nine points smaller than it was.
Every source now declares its tax treatment explicitly, the ex-tax price is resolved once at collection, and comparisons read only that column.
A general-purpose assistant gives a different answer
On 20 August 2026 we asked a chat assistant what a given reference cost in Japan and the United States. It reported the two markets as roughly level. Measured the same day on a like-for-like basis, the American market was 17% above the Japanese one.
Both of the assistant's figures were correct. One was a trade-in price and the other a market index, and nothing in the answer distinguished them. The sign of the difference flips, and the pricing of an entire inventory turns with it.
Counting one absence as a sale sells 188 watches in nine hours
A listing leaving a results page is the closest public proxy for a sale. The first implementation reported 188 items "gone" in 9.3 hours — 43 of them from the company's own 778-item catalogue. The ranked pages had simply reordered.
A disappearance is now counted only after two consecutive absences, and the count resets the moment a listing reappears.
Thirty-one checks before anything is published
Every one of those errors produced a plausible number without raising an error. So the assumption that a person would notice was abandoned, and each failure became an executable check.
No report is generated unless 31 checks pass. They ask whether accessories are being counted as watches — a ¥6,350 screen protector for a Datejust 41 once produced an 877× spread — and whether a reference suffix was silently dropped: 228238A treated as 228238 merged a ¥12.8M diamond variant with the ¥8.8M plain one.
When a check fails on correct behaviour, the check is fixed, not relaxed. Three of them did exactly that, rejecting valid data; all three were corrected rather than switched off.
What transfers
Any category with model numbers, the same item listed across several venues, and prices that differ by country fits the same shape. The only part specific to watches is the list of sources.
What this looks like in other industries
The sections below are not measurements. We have not yet run the study above in these industries, so there are no numbers here. What is written is only what you would watch, what decision changes, and where the mistake is.
Inventing plausible results would make this a better read. But this company sells not putting out numbers without an origin. Doing it on our own case page would take the measured study above down with it. These get replaced with real measurements as they arrive.
Consumer electronics and PC accessories
What you watch — three or four marketplaces plus the direct site. Listed price for the same SKU, effective price after loyalty points, shipping, stock.
What changes — whether to cut price. Instead of a weekly across-the-board cut, you move only the SKUs you are actually losing.
Where it goes wrong — points and coupons mean the same listed price can be a different real price. Compare listed prices only, and you lose while believing you are ahead. The reward rate is on the page, so the effective price lives in its own column.
Cosmetics and supplements
What you watch — authorised sellers, plus marketplace, auction and cross-border listings. Seller name and price.
What changes — when a listing drops below the recommended price, you get a read on which channel it is leaking from. Seller names and quantities over time are the signal.
Where it goes wrong — grey-market and authorised stock sit in the same search results. Count a different size, an old formulation, or an overseas variant as the same product and the market reads cheaper than it is. You need the barcode and the size, not the product name.
Apparel and sneaker resale
What you watch — domestic and overseas resale platforms. Recent trade range and listing count, per size.
What changes — whether to buy, and at what price to list. Each size is its own market, so an item-level average decides nothing.
Where it goes wrong — a listing that disappears was not necessarily sold. A withdrawal and a sale are the same shape. You need exactly the rule from the study above: not counted as sold until absent twice running.
Automotive parts and industrial supplies (B2B)
What you watch — distributor and trading-house catalogues, overseas distributors, published tender results.
What changes — supplier negotiation, and sanity-checking a quote. "This seems above market" becomes a named seller and a number.
Where it goes wrong — listed prices change by quantity band, and much of the market says "contact us". Not turning an absence into a number matters more here than anywhere: a blank is published as a blank.
Food and drink going overseas
What you watch — shelf prices for your products and your competitors' at local retail and local e-commerce in the target market.
What changes — the local wholesale price, and whether to export at all. What something sells for over there is not visible from here.
Where it goes wrong — currency and local tax mean the same product changes from cheaper to dearer within a month. Without a record of which rate was applied as of when, last month's deck and this month's cannot be compared.
Beyond these: if the same item appears on several surfaces under a code or a name, and the price differs by country or channel, the shape carries over unchanged. Get in touch to try it on your own catalogue.
Last updated