Foothold Japan — Published AnalysisFHJ · Teardown #02 · July 2026
Japan Content Teardown #02

You don't need to read Japanese to see this one

Disclosure: This teardown analyses "Flowvenza" — a fictional, illustrative composite of a global automation / iPaaS platform, built from patterns common across overseas B2B SaaS brands operating in Japan. No real company is named, and no real company's copy is reproduced: every Japanese sample below is original writing by the author, representative of the machine-translated voice these brands tend to ship. So are the product and partner names — "Flowvenza" and every tool it "connects to" are inventions. The intent is constructive — the platforms this composite is drawn from are world-class, and the pattern is an industry-wide one, not any single company's failure. This piece looks only at the gap between translated Japanese and Japan-native Japanese: a gap shared by most overseas SaaS brands operating here.

Teardown #01 looked at a subtle problem: Japanese that is grammatically correct but carries an English heartbeat. You needed to read Japanese to feel it.

This one is different. Picture a typical overseas automation platform in Japan — a category leader everywhere else, thousands of integrations, an AI-agent story, and a Japanese site produced almost entirely by machine translation with no native pass. The result is a rarer and more instructive failure mode: breakage you can see without reading a word of Japanese.

That is why this teardown is ordered the way it is. If you are the HQ-side decision-maker — the person who approved the localisation line item but cannot audit its output — Findings 01 and 02 are for you. They involve the two things on your Japanese pages you can verify from your desk: your own product names, and your own numbers. If those are broken, assume the prose around them is too.

Four findings follow — illustrated through Flowvenza, our composite, each with original rewritten Japanese samples, because analysis without execution is cheap.

Summary of findings
F-01Product names don't survive machine translationLock a proper-noun glossary; translate around names, never through them
F-02The numbers don't survive eitherConvert — currency, separators, counters — into the format Japanese approval documents run on
F-03The half-translated siteFinish the money pages to 100%; a smaller fully-native site beats a larger half-native one
F-04Testimonials translated word-for-word turn eerieRebuild social proof as Japan-style case studies: numbers and workflow context, not enthusiasm
FINDING 01

Your product names are being translated into words

For an integration platform, partner and product names are not decoration — they are the catalogue. And proper nouns are precisely what machine translation is worst at. A representative pattern from our composite's Japanese pages:

「テッサヴォックス台本と連携して、コードレビューを自動化。」
「接続先: テッサヴォックス台本」
「次のプロンプトをTessavox Scriptに貼り付けてください。」 Flowvenza Japan site — representative integration-page copy (illustrative composite)

What happened here: the composite's (fictional) AI coding-assistant partner, "Tessavox Script", has been split in half by the MT engine — "Tessavox" transliterated into katakana, "Script" translated into 「台本」, which means a screenplay: the script an actor reads. So the page invites Japanese developers to connect their code reviews to Tessavox: The Stage Play. Three paragraphs later, the same product appears untranslated as "Tessavox Script" — two renderings of one product on one screen, neither of them consistently searchable.

Why this costs more than a laugh:

  • It breaks search. A Japanese user evaluating the integration googles 「Tessavox 連携」 — the Latin-script name plus a Japanese keyword, which is how Japanese practitioners actually search for foreign tools. A page that says 「テッサヴォックス台本」 ranks for a query nobody will ever type.
  • It breaks trust at the exact moment of intent. The reader on an integration page is your highest-intent visitor. A mangled proper noun tells them: no human who knows this product has read this page.
  • It's inconsistently inconsistent. Elsewhere on the same composite site, some partner names get katakana-ised (「フォーマスイート」) while others stay in Latin script (FormaSuite's neighbours on the same list). Each variant is a separate small signal; together they read as unattended.
Exhibit 1 — Proper nouns, repaired
Before — shipped by MTAfter — Japan-native
テッサヴォックス台本と連携して、コードレビューを自動化。Tessavox Scriptと連携し、コードレビューを自動化。
接続先: テッサヴォックス台本連携先: Tessavox Script

Why this works: Japanese B2B convention keeps foreign product names in their original Latin script — Slack is Slack, not スラック, on every serious Japanese SaaS page. The fix is structural, not stylistic: a locked glossary of every product, feature and partner name, enforced before any sentence-level polish. Translate around the noun; never through it.

FINDING 02

The numbers your buyers need are broken

Japanese enterprise buying runs on a written internal-approval document — the 稟議書 (ringi-sho) — and that document runs on numbers. Your Japanese page is where your internal champion goes to copy those numbers. Here is what our composite gives them:

「4 、 500以上のアプリと連携します。」
「自動化により、潜在的な収益が約$2百万ドル増加しました。」
「今すぐ自動化すべき7こと」
「1。トリガーを選択します。」 Flowvenza Japan site — representative stats and list copy (illustrative composite)

Four distinct number injuries in four lines, and — this is the point — you can see every one of them without reading Japanese:

  • The thousands separator became a punctuation mark. "4,500" shipped as 「4 、 000」-style spacing: the comma was auto-converted to 「、」, the Japanese comma between clauses, with stray spaces. To a Japanese reader this looks like the number was typed by something that has never seen a number.
  • "$2 million" was transliterated, not converted. 「$2百万ドル」 stacks a dollar sign and the word "dollars" around a unit construction (「2百万」) that Japanese doesn't naturally use — native copy counts large numbers in 万 and 億. And a revenue claim left in dollars quietly tells the reader this page wasn't priced for their market.
  • The counter word collapsed. "7 things" became 「7こと」 — Japanese requires a counter (「7つ」), and its absence is roughly as jarring as "7 thing" in English.
  • Even the list numbering broke. "1." auto-converted to 「1。」 — a full stop swapped for a Japanese sentence-ending period, in an ordinal.

The strategic cost: the one part of a foreign-language page an HQ executive can audit is the digits — and the digits are exactly what the Japanese approval process needs intact. If your champion cannot lift a single clean figure from your page into their ringi document, your page has opted out of the Japanese buying process.

Exhibit 2 — Numbers, repaired
Before — shipped by MTAfter — Japan-native
4 、 500以上のアプリと連携します。4,500以上のアプリと連携。
自動化により、潜在的な収益が約$2百万ドル増加しました。自動化により、年間約3億円(約200万ドル)の増収余地を創出。
今すぐ自動化すべき7こと今すぐ自動化すべき7つの業務
1。トリガーを選択します。1. トリガーを選択する

Why this works: the rewrite converts rather than transliterates — yen first with the dollar figure retained in parentheses for credibility, 万/億 units, restored counters, and list formatting a Japanese reader can paste straight into an internal document. Numbers are the cheapest trust you will ever buy in this market. They are also the cheapest trust to destroy.

FINDING 03

The half-translated site

Our composite's Japanese navigation, reconstructed:

料金プラン | テンプレート一覧 | An agent for every workflow | 連携アプリ | Build agents that work while you don't | お客様の声 Flowvenza Japan site — representative navigation and section headers (illustrative composite)

The pattern: the legacy pages — pricing, templates, integrations — were translated in an earlier localisation push, but everything shipped since (typically the newest, most strategic content: the AI-agent line) rides in from the global CMS in English. The Japanese site is perpetually 80% Japanese.

Here is the uncomfortable arithmetic: "every page is 80% Japanese" is not 80% as good as a Japanese site. It is 0% as trustworthy. Because the English isn't distributed randomly — it sits exactly on the newest features, Japanese readers draw the accurate conclusion that the things this vendor currently cares about are not available in my language. And they draw a second, more damaging inference about the future: this is what the support experience, the documentation, and the contract process will feel like after purchase. In a market where buying a SaaS product means defending it to a committee, a mixed-language page is not incomplete content; it is evidence against the purchase.

The fix is triage, not heroics:

  • Finish the money pages to 100%. Homepage, pricing, security, the top product line, and whatever page your Japanese ad spend lands on. No English strings, including navigation, buttons, form labels and cookie banners — chrome is where "almost done" is most visible.
  • Fence off the rest. Pages that can't be maintained in Japanese should link out to the English site explicitly (「英語版ドキュメントへ」) rather than ship half-mixed. Japanese buyers handle "this part is English" gracefully; what they don't forgive is not being able to predict which language the next paragraph will be in.
  • Put the newest product line first in the queue, not last. The current pipeline localises in chronological order of page creation — meaning the flagship AI story is the least Japanese thing on the site. Invert it.

A smaller, fully-native Japanese site outsells a larger, half-native one. Every time.

FINDING 04

Testimonials translated word-for-word turn eerie

English-language B2B social proof runs on casual hyperbole — it's how enthusiasm is performed in that culture. Run it through literal translation and something strange happens. A representative composite testimonial, English original first:

"Honestly, calling Flowvenza a lifesaver is putting it mildly — without it we'd be toast."

And the Japanese the MT pipeline ships:

「正直に言って、Flowvenzaを『命の恩人』と呼ぶのは控えめな言い方です。それがなければ、私たちはトーストになっていたでしょう。」 Flowvenza Japan site — representative customer quote (illustrative composite)

A Japanese reader gets a sentence in which a software vendor is credited as a saver of lives, understatedly, by a customer who would otherwise have become toast — read literally, because the idiom doesn't exist in Japanese. The tone lands somewhere between baffling and faintly menacing. This is not a translation-quality problem a better engine fixes; it is a genre problem. The source text is written in a register Japanese B2B communication doesn't use.

Japanese social proof has its own highly developed genre: the 導入事例 (case study), and its currency is not enthusiasm but precision — the workflow before, the workflow after, the number in between, and the job title of the person saying it. Here is the same customer, rewritten as Japanese buyers expect to meet them:

Exhibit 3 — Social proof, rebuilt for the genre導入事例

「導入前は、受注1件の処理に約40分かかっていました。現在は5分です。月間およそ120時間の工数削減となり、繁忙期の残業がほぼなくなりました。」

— 製造業・営業管理部門の責任者(導入事例より)

"Before Flowvenza, processing one order took about 40 minutes. Now it takes five. That's roughly 120 hours saved per month — and overtime in our busy season has almost disappeared."

Why this works: it is quieter than the English original and more persuasive in Japanese. Forty minutes to five is a claim a committee can interrogate; "we'd be toast" is not. Note what this implies operationally: the highest-trust content format in this market cannot be produced by translating your existing testimonials at all. It has to be produced the way Japanese vendors produce it — a structured interview with the customer, in Japanese, about their approval process, their internal resistance, and what actually changed. Which is, transparently, the service behind this analysis. And so —
ABOUT THIS SERIES

Want your content torn down — constructively?

Japan Content Teardown is an ongoing series analysing how global SaaS brands show up in Japanese, by Kodai Sugo — a Japanese content marketer with 1,500+ articles planned, written and shipped, working with overseas B2B brands as Foothold Japan. Everything in this issue — the glossary, the number conventions, the triage of what to finish first, the interview-driven case studies — is what the Foothold Sprint does: from $500/month, month-to-month, no long-term contract.