KasuBook

1. One-liner

Turns a frontline staffer’s 90-second phone report into the customer-harassment response record Japan’s October 2026 law demands.

2. Trend signal — why now?

On 4 June 2025 Japan’s Diet passed an amendment to the 労働施策総合推進法 (Act on Comprehensive Promotion of Labour Policies) that converts customer-harassment countermeasures from a best-effort suggestion into a 措置義務 — a statutory employment-management obligation. It takes effect 1 October 2026. Five weeks from today.

The scope is the part that matters. There is no SME carve-out and no headcount floor. Every employer in Japan — including a business with a single employee — is in scope from day one. That is a different animal from the usual Japanese labour reform, which grandfathers 中小企業 for two or three years. Here, the corner shop and the trading house land on the same date.

MHLW published the implementing guidelines on 26 February 2026. They spell out five required measures: publish a policy, stand up a consultation desk (相談窓口), promptly and accurately confirm the facts when an incident occurs (迅速かつ正確に確認する), document and retain the case, and set out deterrence steps up to service refusal and police notification. Measures one and two are documents you write once. Measures three and four are a running operational obligation — a thing you must do every time a customer screams at a cashier, forever.

Enforcement is 行政指導 escalating to 公表 — the Minister can name you publicly. In Japan, 公表 for a labour violation is not a fine, it is a recruiting and franchise problem. Separately, the September 2023 revision to the 精神障害の労災認定基準 added “suffering egregious nuisance behaviour from customers, clients or facility users” as a recognised stressor event. So the same incident that triggers the compliance duty also now feeds a 労災 claim and a potential 安全配慮義務違反 damages suit — where the employer’s defence is entirely a question of what it recorded and what it did.

Demand is not speculative. UAゼンセン’s third customer-harassment survey (fielded Jan–Mar 2024, published September 2024) found 46.8% of surveyed service-sector members had been subjected to nuisance behaviour, with 暴言 at 39.8% and 威嚇・脅迫 at 14.7% of most-memorable incidents. That is roughly one in two frontline workers with a live incident to report.

Provenance:

• Signal 1 (demand): 2026年10月1日 カスハラ対策義務化 — all employers incl. sole proprietors, five required measures, no SME exemption; 事実関係の確認 and 記録 are ongoing operational duties — https://www.businesslawyers.jp/articles/1457 — 2026-08-25
• Signal 2 (economic): Incumbent カスハラ services price at ¥5,500–¥30,000/mo for 外部相談窓口 alone; 社労士 sell manual-drafting as separate one-off engagements; a documented budget line already exists — https://resus.jp/topics/2614/ · https://thefirm.jp — 2026-08-25
• Signal 3 (feasibility): The 2026 tool market has organised into 自動応答 / 記録 / コンサル・研修, with 記録 solutions arriving as telephony add-ons and a ¥-priced smartwatch launching 1 Oct 2026 — i.e. vendors are selling capture hardware, not the response record — https://kigyolog.com/service.php?id=787 · https://prtimes.jp/main/html/rd/p/000000021.000054401.html — 2026-08-25
• Signal 4 (economic, second-order): Sept 2023 労災認定基準 revision added customer nuisance behaviour as a recognised psychiatric stressor, exposing employers to 安全配慮義務違反 damages where records are absent — https://law-bright.com/corporationlaw/contents/customer-harassment/industrial-accident-certification/ — 2026-08-25
• Signal 5 (demand, quantitative): UAゼンセン 2024 survey — 46.8% of service-sector respondents experienced nuisance behaviour; 暴言 39.8%, 威嚇・脅迫 14.7% — https://uazensen.jp/2025/06/04/121677/ — 2026-08-25 Category: Regulatory arbitrage (a compliance duty landing on a newly-in-scope population with no size exemption) + Underserved niche (vendors sell capture hardware and consultation desks; nobody sells the response record the inspector actually asks for)

3. The opportunity

The Japanese カスハラ vendor market has organised itself into three buckets — 自動応答 (auto-response / IVR), 記録 (capture), and コンサル・研修 (consulting and training). Every one of them is selling the wrong half of the obligation.

The 記録 vendors sell capture: call recording bolted onto a cloud PBX, an AI transcription add-on, a smartwatch that starts recording when it hears a raised voice. All of it is aimed at getting audio. But the guideline does not require audio. It requires that when an incident occurs the employer promptly and accurately confirms the facts, documents the case, and can show what it did about it. Audio is evidence input. The obligation is about the output — a dated case record showing who was involved, what happened, who at the company was told, what was decided, and whether the worker was followed up with.

The コンサル vendors and 社労士 sell the opposite end: a beautifully drafted 対応マニュアル and a 就業規則 clause. That is measure one. It is a document you buy once, in September 2026, and then it sits in a binder while your staff keep getting shouted at.

Nobody is selling the middle. And the middle is where the failure actually happens, for a completely mundane reason: the person who witnesses the incident is a part-time cashier at 19:40 on a Saturday, and the record has to be written by them, from memory, in Japanese, into a form. It never gets written. The manager hears about it verbally, sympathises, and moves on. Six months later a 労働基準監督署 inspector asks how many incidents you handled and what you did, or a former employee’s lawyer asks the same question with damages attached, and the honest answer is “we don’t have that.”

The 10× is not analysis. It’s capture friction. If reporting an incident is a phone call — the staffer talks for ninety seconds in whatever Japanese comes out, or in Portuguese or Vietnamese if that’s what they speak — and the structured 対応記録票 writes itself, incidents get recorded. If it’s a form, they don’t. Everything downstream (aggregation, the annual summary, the inspector packet) is trivial once the records exist. They currently don’t exist.

Second-order: Japan’s frontline retail and food-service workforce is heavily non-Japanese — 特定技能 and 技能実習 staff in konbini, family restaurants, hotels and care homes. Those are exactly the workers most exposed to 暴言 and least able to write a formal Japanese incident report. A capture layer that accepts spoken Vietnamese, Nepali, Indonesian, Burmese or Portuguese and emits correct Japanese business prose is not a feature — for a large slice of the target market it is the entire product.

4. Target market

• Primary customer: 総務 / 人事 manager, or the owner-operator, at a Japanese multi-site service business with 20–300 employees across 3–40 locations — regional supermarket and drugstore chains, family-restaurant and izakaya groups, business hotels and ryokan, ward-level care providers (介護事業所), driving schools, clinic groups, and multi-branch 士業 or municipal contract operators. Annual revenue roughly ¥300M–¥10B. Crucially: too big to hand-wave the obligation, too small to have a compliance department.

• Why they buy: They already know 10/1 is coming — the 社労士 told them, the industry association newsletter told them, their insurer told them. They have or will buy a manual. What they do not have is any answer to “and then what happens when it actually occurs at store 7?” Their honest current state is that the store manager LINEs the head office, someone reads it, and nothing is written down. They are buying the ability to say, to an inspector or to a lawyer, here is the case, here is the date, here is what we did.

• Rough TAM reasoning: Japan has on the order of 3.3–3.6 million SMEs, but the addressable slice is multi-site, customer-facing, and large enough to have a designated 総務 function — realistically low hundreds of thousands of businesses across retail, food service, lodging, care, and personal services. At a ¥19,800/mo entry point, capturing 500 of them is ¥119M ARR ($790K). Capturing 2,000 is ¥475M ($3.2M). No unicorn TAM required, and that is fine.

• Why now for them: A hard statutory date five weeks out, with no size exemption, applying to a business function they have never systematically documented. The budget conversation is already happening — it is happening this quarter, for this fiscal year, in the 総務 budget, next to the manual quote from the 社労士.

5. Product sketch (MVP)

• Phone-in incident capture. A dedicated number per company. Staff call, speak for 60–120 seconds in plain speech, hang up. No app install, no login, no form — works on a personal phone, works for a 70-year-old part-timer.
• Auto-drafted 対応記録票. The call becomes a structured Japanese incident record: date/time, location, staff involved, customer behaviour classified against the MHLW/UAゼンセン categories (暴言・威嚇脅迫・長時間拘束・繰り返しクレーム・SNS投稿の示唆 etc.), severity, immediate action taken, and open follow-ups.
• Multilingual in, Japanese out. Reports spoken in Vietnamese, Nepali, Indonesian, Burmese, Tagalog, Portuguese or English produce a correct Japanese 記録票 plus the original transcript retained alongside it.
• Escalation routing. Severity thresholds notify the designated 相談窓口 owner and the store’s supervising manager by email/LINE WORKS, with a required acknowledgement — which itself becomes part of the record, evidencing that the employee was not left to face the customer alone.
• Follow-up ledger. Tracks the worker-care step the guideline expects: was the staffer checked on, was 産業医 or the EAP offered, was it closed out. Open items age visibly.
• Repeat-offender linking. Same customer, same store, third incident — surfaces automatically, which is the trigger for the deterrence measures (warning, service refusal, ban, police).
• Inspector packet, one click. Period-scoped PDF: policy version in force, incident counts by category and site, response timelines, and the full case list. This is the artefact for 行政指導, for a 労災 inquiry, and for the annual 安全衛生委員会.
• Attach evidence. Existing call recordings, store camera stills, or the customer’s letter drop onto the case rather than living in someone’s inbox.

6. AI angle — what’s load-bearing

Remove the AI and this is a web form, which is exactly the thing that already exists and already fails. The entire product thesis is that the record only gets written if writing it costs the frontline worker nothing.

Three places AI is doing genuine work:

1. Speech-to-structured-record. A distressed part-timer’s rambling ninety seconds — with store noise, dialect, and no narrative order — becomes a clean, chronologically ordered, categorised Japanese 記録票 in business register. This is the whole ballgame. A human transcription service costs more per incident than the subscription.
2. Cross-lingual normalisation. Spoken Vietnamese or Nepali → correct Japanese 記録票 that a 労基署 inspector will accept as the employer’s own record, with the source-language transcript retained for fidelity. There is no non-AI version of this at ¥20K/mo.
3. Classification and severity triage. Mapping free narrative onto the guideline’s harassment categories and flagging the cases that legally require escalation, plus entity-resolving the same repeat customer across stores and reports where nobody wrote down a name.

Notably the AI is not deciding legal outcomes or judging whether behaviour was harassment — that would be both a liability problem and a bad product. It drafts; the human 相談窓口 owner reviews and confirms. That review click is itself compliance evidence.

7. Localization angle

This is a Japan-only product and could not be anything else. The statute, the MHLW guideline categories, the 記録票 format, the 労基署 inspection ritual, and the 社労士 distribution channel are all Japanese and all load-bearing. The business-Japanese output register (敬体, 事実の記載, no editorialising) is not a translation problem — get it wrong and the record reads as an employee’s complaint rather than the company’s official record.

The genuinely interesting local wrinkle is the inverse one: Japan’s frontline service workforce is increasingly non-Japanese-speaking, so a Japan-only compliance product needs to be aggressively multilingual on the input side. That combination — foreign-language in, formal Japanese out, mapped onto a Japanese statutory schema — is a wedge no global HR-tech vendor will build and no domestic 社労士 tool will either.

Pricing must be yen-native (¥19,800 / ¥49,800 / ¥98,000 tiers read as normal SME software; $199 does not), and billing must support 請求書払い with bank transfer, not just card — most Japanese SMEs will not put a recurring subscription on a corporate card.

8. Business model — path to $1M–$5M ARR

• Pricing: ¥19,800/mo up to 5 locations · ¥49,800/mo up to 20 locations · ¥98,000/mo up to 50 locations, then enterprise. Priced per site, not per seat — every frontline worker must be able to call in, so per-seat pricing would actively destroy the product.
• ACV: ~¥240K–¥600K ($1.6K–$4K). Blended target ¥340K ($2.3K).
• Rough math to $1M ARR: ~440 customers at ¥28K/mo blended = ¥148M ≈ $985K. Reachable with a single 社労士 channel plus direct outbound to regional chains.
• Rough math to $5M ARR: ~1,800–2,200 customers, or ~1,200 customers with a healthier mix skewed toward the ¥49,800 and ¥98,000 tiers. Requires either a genuine 社労士 partner network (30–50 firms actively reselling) or one industry-association endorsement — supermarket, food-service, or care-provider association — that puts the product in front of thousands of members at once.
• Expansion path: Sites is the natural meter and it grows with the customer. Beyond that: (a) extend the same capture spine to 従業員間ハラスメント (power/sexual harassment), which is already a 措置義務 since 2020 and has the identical record-keeping gap; (b) a 産業医/EAP referral module; (c) an annual benchmark report — “your 暴言 rate vs. peers in your prefecture and sector” — which is a paid upsell and a defensibility play at the same time.

9. Go-to-market wedge — first 100 customers

1. 社労士 channel, built October–December 2026. Roughly 45,000 registered 社会保険労務士 in Japan; a large number are right now selling カスハラ manual-drafting as a one-off engagement and have nothing recurring to follow it with. Target the ~800 firms with a visible corporate-client web presence and a published カスハラ page (they self-identify by writing the blog post). Pitch: you keep the manual fee, we pay 20% recurring on the tool, your client stops calling you every time an incident happens. Realistic: 60 conversations → 20 partner firms → 2–4 clients each → 40–80 customers.
2. Industry-association seminar circuit. Prefectural supermarket associations, 商工会議所, food-service and care-provider associations all run compliance seminars, and カスハラ is the 2026 topic. Speaking slots for a practical “what an inspector will actually ask you for” session are gettable for free because the associations need speakers. One seminar, 40–80 attendees, all in-scope, all with the deadline in their calendar.
3. Direct outbound to regional multi-site chains, ranked by exposure. Build a list of 2,000 regional supermarket, drugstore, family-restaurant, business-hotel and 介護 operators with 3–40 sites from 帝国データバンク / prefectural directories. Send a physical letter (still works in Japan for 総務, and cuts through where email doesn’t) containing a filled-in sample 記録票 for a plausible incident in their industry. Expect 3–5% response; 60–100 conversations.
4. The UAゼンセン-affiliated angle. Union-affiliated retail and service employers are under direct member pressure to show they are doing something concrete; the union has publicly campaigned for this law and its members will be asking. Employers in that position want an artefact, not a promise.
5. Deadline-timed content in Japanese. Not generic SEO — one specific asset: a free, accurate, downloadable カスハラ対応記録票 template matched to the February 2026 guideline, gated on email. Everyone searching for that template in September is a buyer with a deadline.

10. Build complexity — justification

Low. The stack is off-the-shelf: a telephony provider for the inbound number and recording, a speech model handling Japanese plus the relevant Southeast Asian languages, an LLM for structuring and classification, and a standard web app for the case list, review queue and PDF export. No custom models, no hardware, no integrations required for v1 — the escalation notification can be email and LINE WORKS webhook, both trivial.

The real work is not engineering, it is schema and register fidelity: getting the 記録票 fields, the harassment taxonomy, and the Japanese business prose exactly right against the MHLW guideline, which needs a 社労士 or employment lawyer reviewing output for several weeks. Budget 8–10 weeks to a v1 a real customer can run, with a domain reviewer on retainer. A technical solo founder with Japanese-language capability, or a two-person founder pair where one is Japanese-native, ships this.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Helps employers meet a statutory duty. Recording and personal-data handling must comply with 個人情報保護法 — the guideline explicitly contemplates recording subject to that. No licensed activity: the product drafts records, it does not give legal advice.
Ethical — no harm / dark patterns	✅	Protects frontline workers, and improves the evidence base for their own 労災 claims. One genuine tension: the same record can be used against a worker. Mitigation — workers see their own reports, and the product does not adjudicate.
Market exists (evidence above)	✅	Statutory duty with a hard date and no size exemption; incumbents already charging ¥5,500–¥30,000/mo in adjacent slots.
1–5 person team can build this	✅	Off-the-shelf speech + LLM + web app. Domain reviewer on retainer, not headcount.
Launchable with <$50K / ₹40L	✅	Telephony, inference and hosting are usage-priced; the meaningful spend is the 社労士 review retainer. Well under $50K.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Hard statutory date, no exemption, 公表 risk, and a second exposure path via 労災 / 安全配慮義務. Not a full 17–20 because a non-compliant employer’s felt pain stays low until an inspection or a claim — the pain is real but latent for most buyers most of the time.
Demand evidence	15	12/15	Strong: legislated duty, published guideline, priced incumbents in adjacent slots, quantified incident prevalence (UAゼンセン 46.8%). Held below 13 because the demand evidence is for カスハラ compliance spend generally, not specifically for a record-keeping tool — I have not yet found employers saying “we can’t get incidents written down.”
Build feasibility	13	13/15	Off-the-shelf everything; 8–10 weeks. Docked for the schema/register fidelity work, which is genuinely fiddly and needs paid domain review.
Distribution clarity	15	11/15	Named channels with countable lists (社労士 firms, association seminars, regional chain directories) and a plausible conversion path. Docked because 社労士 partner channels in Japan are relationship-slow, and a foreign or unknown founder faces a trust gap that costs weeks per partner.
Revenue mechanics	15	11/15	Per-site pricing is correct and benchmarked against ¥5,500–30,000/mo incumbents. $1M needs ~440 customers, which is a lot of small logos for a small team; $5M needs a real channel or an association deal that isn’t yet in hand.
Time to first revenue	10	8/10	A pre-sold pilot before 1 October is realistic given the deadline. Not 9–10 because Japanese SME procurement rarely closes in under 4–6 weeks even when the buyer is motivated.
Defensibility	10	4/10	Honestly weak. Any competent Japanese vendor can copy this in a quarter, and the incumbent 記録 vendors have telephony distribution already. The only accumulating moats are the incident corpus (benchmarking) and 社労士 relationships. Execution and timing moat, not a structural one.
Total	100	75/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

Japanese-language capability is not optional — it is a hard requirement for both the product output and the 社労士 channel. A non-Japanese-speaking founder should not attempt this alone.

Key assumptions to validate (3–5)

1. Assumption: The record-keeping gap is real — multi-site employers currently have no systematic written record of customer-harassment incidents. How to test: 20 structured interviews with 総務 managers at 3–40-site chains. Ask one question: “how many customer-harassment incidents did you have last quarter, and can you show me the records?” If more than half produce records, the wedge is gone.
2. Assumption: Frontline staff will actually phone in a report, and part-timers will do it without a manager standing over them. How to test: Run the capture flow manually for 2 weeks at one friendly 5-site chain (human transcription behind the curtain). Measure reports per site per week. Below ~0.3 and the product has no data to work with.
3. Assumption: 社労士 firms will resell rather than build a spreadsheet template and keep the whole fee. How to test: Pitch 15 firms with the partner economics. Look for signed intent, not enthusiasm.
4. Assumption: ¥19,800–¥98,000/mo clears the 総務 budget without a lengthy 稟議. How to test: Ask directly in the interviews — what is your approval threshold, and who signs at this number.
5. Assumption: The AI-drafted 記録票 is accepted as the employer’s own record and reads correctly to a 社労士. How to test: Have three independent 社労士 blind-review 20 generated records against the February 2026 guideline.

Risk flags

1. Competitive timing: The incumbent 記録 vendors (cloud PBX, call recording, the smartwatch launching 1 Oct 2026) sit one product decision away from adding the record layer, and they already own the distribution. The window is the twelve months around the deadline.
2. Latent-pain risk: This is compliance software for an obligation whose enforcement is 行政指導 rather than an automatic fine. Some employers will rationally decide a manual in a binder is enough and buy nothing. Churn after the deadline passes without an inspection is a genuine threat — the annual benchmark report and the harassment-module expansion exist partly as retention answers.
3. Data sensitivity: Incident records contain named employees, named customers, and mental-health-adjacent information. A breach is existential in Japan in a way it is not elsewhere. 個人情報保護法 handling, retention limits, and access control need to be right at v1, not v2.
4. Channel trust gap: 社労士 partnerships and association speaking slots reward incumbency and introductions. A cold outsider — especially a non-Japanese one — should expect the channel to take twice as long as the model assumes.
5. Weak defensibility: 4/10 is the honest number. This is a speed-and-focus business with a twelve-month window, not a durable moat.

14. Structured verdict

Score:                  75/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Japanese-native (or fluent) technical founder, solo or pair,
                        with a 社労士 or employment lawyer on retainer for schema review
Time to revenue:        6–10 weeks (pre-sold pilot ahead of the 1 Oct 2026 deadline)
Capital to launch:      ¥2–4M (~$14–27K), mostly domain-review retainer and inference
Top 3 assumptions to validate first:
  1. Record-keeping gap is real — 20 総務 interviews; ask them to show last quarter's records
  2. Frontline staff will phone in reports — 2-week manual pilot at one 5-site chain, measure
     reports/site/week (kill below 0.3)
  3. 社労士 firms will resell — pitch 15 firms, look for signed intent on 20% recurring
Kill criteria:
  - Abandon if >50% of 20 interviewed 総務 managers can produce existing incident records
  - Abandon if the 2-week manual pilot yields <0.3 reports per site per week
  - Abandon if an incumbent 記録 vendor ships an equivalent response-record layer before v1,
    given they already hold the telephony distribution
  - Abandon if fewer than 3 of 15 pitched 社労士 firms will commit to a partner arrangement

15. Next step — 1-week validation sprint

• Day 1–2: Build the list. Pull 60 regional multi-site operators (supermarket, drugstore, family restaurant, business hotel, 介護) with 3–40 sites, plus 15 社労士 firms with a published カスハラ page. Draft the interview script around the single killer question: “How many customer-harassment incidents did you have last quarter, and can you show me the record?”
• Day 3–4: Run 20 interviews with 総務 / owner-operators. Do not pitch. Record three things per interview: (a) can they produce records, (b) what happens today when an incident occurs at a remote site, (c) what number would need a 稟議 rather than a manager’s signature. In parallel, pitch the 15 社労士 firms on the partner economics.
• Day 5: Generate 20 sample 記録票 from realistic incident narratives and have two 社労士 blind-review them against the February 2026 guideline. Decide.

Falsifiable go/no-go: Proceed only if (a) ≥14 of 20 interviewed employers cannot produce any written incident record for the last quarter, (b) ≥3 of 15 社労士 firms verbally commit to a partner arrangement, and (c) ≥16 of 20 generated 記録票 are rated by both reviewers as acceptable as an employer’s own record with only minor edits.

Fail any one of the three and this is a VALIDATE, not a GO — and given the defensibility score, a VALIDATE that misses the October window is a PASS.