MitsuMoru — LINE photo→見積書 AI for Japan’s solo リフォーム

1. One-liner

LINE bot that turns jobsite photos + voice memos into a customer-ready 見積書 PDF for Japan’s solo リフォーム contractors.

2. Trend signal — why now?

Japan has a labor crisis the rest of the world doesn’t quite have. Carpenters dropped from 937K (1980) → 298K (2020) → projected 130K by 2040. 25% of solo contractors (一人親方) are 65+. Job-to-applicant ratio is 1.18, the BoJ Tankan diffusion index for employment fell to -35 — three-decade low. There are 11M missing workers projected by 2026. Solo trades can’t hire their way out, so the only lever left is automating the office work that eats their nights.

And the office work is brutal. A standard estimate practice piece states a ¥10M project takes ~15 hours of estimate work. Reform-industry blogs describe owners doing fieldwork all day, then estimate paperwork at night. ANDPAD/アイピア-class ERPs cost ¥10K-30K+/mo and are aimed at companies with dedicated office staff — not the one-person shop with a beat-up Hijet and a smartphone.

Two things shifted in the last 12 months that crack this open:

1. Vision LLMs got cheap and good at Japanese. GPT-4-class vision + Gemini 2 + Whisper-grade JA ASR can now read a tile bathroom, identify the unit bath model class, count linear meters of trim, and write the kitchen-replacement scope in passable Japanese. LIXIL launched ラクみつ in April 2025 — proof the photo→quote flow works — but it’s locked to LIXIL’s own インプラス internal-window product (a single SKU upsell tool, not a general estimator).
2. The IT subsidy retooled around AI. “IT導入補助金 2025” became “デジタル化・AI導入補助金 2026” with a 4/5 subsidy on small-tool spend ≤¥500K — explicitly listing 見積積算 software for construction/reform. 一人親方 and individual contractors are eligible. A ¥3,980/mo tool costs the contractor ~¥800/mo after subsidy in year one. That’s the wedge that puts SaaS into the 一人親方’s price band.

LIXIL ラクみつ proved consumers and pros respond to the photo-AI flow. Nobody has shipped the vendor-neutral, LINE-first, voice-first, generalist version for the solo reform contractor. That’s the gap.

Provenance:
  - Signal 1 (demand): Carpenter workforce 937K (1980) → 298K (2020) → ~130K by 2040; 25% of 一人親方 are 65+; reform owners do estimates at night after fieldwork; ¥10M project = ~15 hrs estimate work — https://souken.craft-bank.com/analisys/chokuei/ + https://aippearcloud.com/estimate_time/ — accessed 2026-05-04
  - Signal 2 (feasibility): LIXIL "L-ポケット ラクみつ" launched 2025-04 — photo→AI estimate works for windows; vision+JA ASR APIs commodity; LINE Messaging API ubiquitous (SMB CR > 90% in JP) — https://newsroom.lixil.com/ja/2025041702 — accessed 2026-05-04
  - Signal 3 (economic): "IT導入補助金 2025" → "デジタル化・AI導入補助金 2026" — 4/5 subsidy on ≤¥500K small-biz tools, 見積積算 explicitly covered, 一人親方 eligible; ANDPAD operates SoftBank-distributed enterprise tier at ¥10K-30K+/mo and is "high price" per public reviews — https://it-shien.smrj.go.jp/about/ + https://www.softbank.jp/biz/services/collaboration/cbo/ — accessed 2026-05-04
  Category: Tech-unlock + Geographic arbitrage

3. The opportunity

The Japanese reform/handyman/一人親方 segment is structurally underserved. The big ERPs (ANDPAD 230K companies, アイピア 500+ companies, Photoruction 400K projects) all chase mid-size 工務店 with office staff. Their UX is web/desktop with mobile bolt-ons. Their pricing assumes a dedicated estimator role that solo shops don’t have. Their AI features (ANDPAD’s recent agents, Photoruction’s KENGI 2026 rollout) are aimed at drawings → quantity-takeoff for new builds, not “I’m standing in a 25-year-old bathroom, the customer wants it gone, write me the quote.”

LIXIL’s ラクみつ proved photo→quote works but is a vendor-controlled single-product upsell tool. 写真de見積もりくん (yamamoto-kun.co.jp) targets exterior painting only. The general-purpose, vendor-neutral, LINE-first flow for the solo operator does not exist.

The wedge is brutal simplicity: contractor adds @MitsuMoru on LINE, sends 4-6 photos and a 30-second voice memo (“浴室全部やりたい、ユニットバス1418、給湯器も交換、来月着工目標”), gets a 見積書 PDF in their LINE thread within 5 minutes that they forward to the homeowner unchanged, or tweak in 90 seconds before sending. No app to install. No web portal to log into. No drawings.

What MitsuMoru does that no incumbent does:

• LINE bot — zero install friction, the channel solo trades already use with customers
• Voice-first input — works on a ladder, in a bathroom, with one hand
• Vendor-neutral product DB — TOTO, LIXIL, Daiwa, Cleanup, KEIYO等 unit-bath/kitchen/window/door catalogs all priced
• Solo-tuned pricing — ¥3,980/mo, deliberately under the IT subsidy ¥500K cap, eligible for 4/5 subsidy
• Customer-ready output — Japanese 見積書 format that satisfies 建設業法 written-estimate rules

4. Target market

• Primary customer: Solo + 1-3 person reform contractors (リフォーム業) and 一人親方 in Japan. Specifically: kitchen/bath/exterior/interior/小規模住宅 reform contractors with annual revenue ¥30M-¥150M, no dedicated office staff, owner-operator handles estimates personally.
• Why they buy: Estimates eat 8-15 hours/week of evening/Sunday time. Win rate suffers because they’re slow to respond to inquiries. Existing tools (ANDPAD, アイピア, Photoruction, AnyONE) cost ¥10K-30K+/mo, require web/PC workflow, and are designed for offices with staff. A solo who pulled in ¥5.21M (the JP avg 一人親方 income) can’t justify enterprise pricing.
• Rough TAM reasoning: ~200K 一人親方 in construction trades (Craft Bank Souken). ~50K-80K of them in the reform/maintenance subset that runs estimates regularly. Adjacent: ~30K 便利屋/handyman ops. ~20K small 工務店 with no dedicated estimator. Realistic addressable: ~100K shops. At 1% penetration × ¥3,980/mo × 12 = ~¥48M ARR. At 5% → ¥240M. Sub-$5M ARR target is reachable on this single segment alone.
• Why now for them: (a) 2026 IT subsidy redesign explicitly funds AI tools at 4/5 — first time that lowers the price floor for solo. (b) Labor shortage is now their #1 lived pain — they cannot hire an admin person. (c) LINE Official Account is already the primary customer channel for ~70% of small reform shops, so the friction to add a bot to their workflow is near zero. (d) Customers (homeowners) are getting 相見積もり quotes from 3-5 contractors via online matching sites (リショップナビ, ホームプロ等) — slow estimate response = lost job.

5. Product sketch (MVP)

• LINE Official Account “@MitsuMoru” — contractor adds it as a friend, links their account in 60 seconds (business name, license number, default tax/profit margin)
• Photo + voice ingestion — drop 1-12 jobsite photos and a voice memo; bot replies with an interpreted scope summary the contractor confirms or edits in chat
• Estimate generation — 5-minute SLA from “scope confirmed” to PDF; itemized 見積書 with materials, labor (人工), 諸経費, 値引き, 消費税, 有効期限
• Vendor-neutral product catalog — TOTO/LIXIL/Daiwa/Cleanup unit baths, kitchens, windows, doors, sash, flooring, roofing, exterior paint; with current 定価 + standard discount tiers
• Personal price book — every estimate the contractor edits trains their personal margins/preferred SKUs; gets faster + more accurate over time
• One-tap PDF send — forward the 見積書 PDF to the customer in the same LINE thread, or download for paper/email
• Win-tracking — bot asks “受注しましたか？” two weeks after; binary feedback feeds the price book
• 建設業法-compliant template — required fields (工事内容、施工場所、工期、支払条件、請負代金) prefilled

Not in MVP: drawings/CAD, 工程管理, 写真台帳, 請求書 (handled by 既存 freee/MFInvoice/INVOY which are universal in this segment).

6. AI angle — what’s load-bearing

AI does the actual work, three places:

1. Vision scope reading. From photos, the model identifies room type, fixture types, approximate dimensions, condition (replace vs repair), accessibility constraints. Without vision LLMs this requires a site visit + measuring + manual itemization — the 15-hour bottleneck.
2. Voice intent parsing. 30-second JA voice memo → structured scope (rooms, products, optional add-ons, target start date, customer constraints). Without ASR + JA reasoning this needs typed forms, which solos won’t fill on a phone in a bathroom.
3. Estimate composition. Mapping scope → product SKUs → quantities → personalized margins → 建設業法-compliant Japanese PDF. Templated SaaS gets you the form; the LLM fills it the way the contractor would on a Saturday night, but in 5 minutes.

Remove the AI and you’re left with a LINE bot that lets you upload a photo and fill a form. That’s not the product. The AI is load-bearing.

7. Localization angle (if any)

This is a Japan-only play. Why:

• LINE penetration in JP small business communication is roughly 2x any other messaging channel; same is not true in any other market
• 建設業法 estimate format requirements are JP-specific (工期、請負代金、支払条件 etc.)
• Japanese product catalog (TOTO, LIXIL, Daiwa, Cleanup) is a closed ecosystem — Western product DBs don’t help
• Subsidy mechanic (デジタル化・AI導入補助金) is JP-specific and is the single biggest CAC reducer
• Japanese voice + Japanese vision text recognition is a model-quality cliff that has only just been crossed

A Korea/Taiwan adaptation is plausible later (KakaoTalk + local product catalogs) but should not influence v1.

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

• Pricing: ¥3,980/mo flat (Solo); ¥7,980/mo for 2-3 user shops (Team). Annual prepay 2-month discount. Deliberately set under ¥500K cumulative cap so a 3-year subsidy claim works.
• ACV: Solo ¥47,760/yr; Team ¥95,760/yr. Blended at 80/20 mix → ¥57K/yr ($365 USD/yr).
• Path to $1M ARR (~¥150M): ~2,600 paying shops at blended ACV. ~2.6% of the ~100K addressable. Believable in 14-18 months given subsidy-funded sales motion.
• Path to $5M ARR (~¥750M): ~13K shops. 13% penetration of 100K. Requires expansion to adjacent verticals (便利屋, 板金塗装, 外構, 設備工事) — each of which is structurally identical (solo, LINE-native, photo-quotable). Plausible 30-month path.
• Expansion path: Add (a) AI 請求書 (post-job invoice) for ¥1,500/mo, (b) AI customer-followup (“3 weeks since estimate sent — should we ping?”), (c) lead-gen referral fees from product vendors (TOTO/LIXIL pay for last-mile contractor recommendation) — each lifts ACV 30-60% without adding new acquisition cost. Eventually a marketplace cut on materials orders.

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

The IT subsidy is the single biggest wedge. Path:

1. Become a registered IT導入支援事業者. This is a paperwork process, not a competition. Once registered, MitsuMoru itself appears in the official ITツール catalog. Every prospect’s natural Google search for “見積 アプリ 補助金” surfaces the product. Cost: ~¥0 + 6 weeks of form-filling. ~7,000 shops/yr apply for this subsidy — tens of % of the catalog browsing audience is target customer.
2. Cold outreach to LINE-active reform shops on リショップナビ + ホームプロ + 外壁塗装パートナーズ directories. Scrape ~5,000 shop names + LINE official account IDs (publicly visible). Send a personalized 90-second demo video via the contractor’s own LINE channel (“ここで動いてるところを見てください”). Conversion target: 3-5%. → 150-250 free trials, 50-80 paid conversions on the first sweep.
3. YouTube + TikTok demos in JA. Channels like 工務店経営塾, 建設業バックオフィスチャンネル, リフォーム職人系インフルエンサー have 30K-200K subscribers each. Sponsored demo + live walk-through of a real 見積 generated in 5 minutes. ¥150K-¥300K per video; 3-4 videos to sequence the funnel.
4. 建設業法 webinar with a 行政書士. Co-host a 1-hour session on “電子見積書 + 建設業法 + 補助金申請 — 一人親方が今やるべきこと.” 100-200 attendees per session, ~10% trial conversion. The 行政書士 also becomes a referral channel for IT subsidy filings.
5. LINE-native referral. The PDF includes a small footer “MitsuMoru で作成 — リフォーム業の見積革命” with an opt-out. 30-50% of homeowners are themselves contractor-adjacent (cousins, parents, neighbors) — passive virality is real in regional Japan.

I can see the first 100 paying customers concretely from channels 1+2 alone, in the 90-day window post-launch.

10. Build complexity — justification

Medium. v1 is: LINE Messaging API webhook + GPT-4-class vision + Whisper-style JA ASR + a 5K-SKU product catalog (semi-public from manufacturer PDFs, scraped/normalized) + a JA PDF generator (jspdf + 建設業法 template) + Stripe + a personal price book db. No custom models, no novel infra. Tricky parts: (a) curating + maintaining the product catalog, (b) personal price-book training loop, (c) producing 見積書 PDFs that pass building-industry expectations on first send. Realistic: 2 engineers, 4-5 months to v1; another 2 months to subsidy registration + first cohort.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	建設業法 covers form/disclosure of estimates; SaaS itself unregulated. No 個人情報保護法 issue beyond standard contractor PII.
Ethical — no harm / dark patterns	✅	Contractor confirms scope before PDF generation. No auto-send to customers without owner approval.
Market exists (evidence above)	✅	ANDPAD/アイピア/Photoruction prove SaaS in this category sells; LIXIL ラクみつ proves photo-AI flow lands; 100K-shop addressable bottom segment.
1–5 person team can build this	✅	2 engineers, 4-5 months.
Launchable with <$50K / ₹40L / ¥7M	✅	API costs + LINE Business Account + a low-res vendor catalog scrape + a part-time JA copywriter for the 建設業法 PDF template. <¥5M to MVP.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Estimate work is the documented #1 evening burn for solo reform contractors; ¥10M project = 15 hrs estimate; lost jobs from slow response. Hair-on-fire weekly.
Demand evidence	15	12/15	Multiple incumbents in adjacent tiers (ANDPAD, アイピア, Photoruction, AnyONE), LIXIL ラクみつ launched in same direction, 7K+ shops/yr already file IT subsidy for these tools. Direct demand for the specific solo-tier flow is inferred, not measured. -3.
Build feasibility	15	11/15	Off-the-shelf APIs but nontrivial: vendor catalog curation, personal price-book training, JA-quality PDF generation, edge cases on photo coverage. 4-5 months realistic.
Distribution clarity	15	12/15	IT subsidy registration + scraped LINE directories + 行政書士 partner + creator sponsorships; channels are named, math is plausible but unproven at scale.
Revenue mechanics	15	11/15	¥3,980/mo undercuts incumbents 3-5x; subsidy puts effective price at ~¥800/mo; 100K addressable shops support $5M ARR. Risk: churn after subsidy year ends.
Time to first revenue	10	8/10	Pre-sale-able to 10-20 friendly contractors during MVP; first paid cohort within 8-12 weeks of soft launch.
Defensibility	10	8/10	Personal price-book per shop = high switching cost after 6 months; vendor catalog + 建設業法 template = a moat against generic global tools (OpenAI Operator, etc.). Subsidy registration is a 2-month wall for fast-followers.
Total	100	79/100	GO.

13. Qualitative modifiers

Founder-fit tags

technical-heavy (vision + ASR + JA NLG plumbing must be solid) and domain-expertise-required (must understand 建設業法 estimate rules, JP construction product catalog, the actual reform workflow). Best fit: a Japanese (or JA-fluent) engineer-founder paired with a 元工務店 advisor. Sales-heavy is not required — IT subsidy + LINE-native distribution does most of the lifting.

Key assumptions to validate (3–5)

1. Assumption: Solo reform contractors will trust an AI-generated 見積書 enough to send to homeowners with light edits (90 sec or less). How to test: 20 shadow runs with paying contractors during a 4-week pilot; measure edit-time-before-send and re-send rate.
2. Assumption: The 5-min SLA on vision+voice→PDF is achievable at <¥80 per estimate at v1 scale (~10K estimates/mo). How to test: Cost spreadsheet using current GPT-4-class + Whisper rates; live timing on 50 staged inputs.
3. Assumption: Subsidy-eligibility moves the close rate by 2-3x vs. an unsubsidized SaaS pitch. How to test: Run two cold-outreach arms in week 1 (subsidy-led vs. value-led messaging) on 200 prospects; measure trial-start rate.
4. Assumption: Year-2 churn after subsidy lapses stays below 30%. How to test: Cannot validate pre-launch; build kill criterion around it (below).

Risk flags

1. Platform dependency: LINE Messaging API. LINE has changed pricing/limits before (notably 2023 free-tier cuts). Mitigation: support a web upload alongside LINE from day one, even if it’s the secondary path.
2. Vendor catalog maintenance: TOTO/LIXIL/etc. update SKUs and pricing quarterly. Manual maintenance grinds margin. Mitigation: scrape + LLM-normalize, but expect 1 part-time ops headcount by month 6.
3. Regulatory drift: 建設業法 updates (recent: 2024 revisions on subcontractor 見積 disclosure) could add fields. Mitigation: 行政書士 advisor on retainer + monthly template review.
4. AI accuracy on bad photos: Solo contractors take dark, partial shots. Mitigation: bot prompts for missing angles before generating; fallback to text-only mode preserves the value prop even on photo-only-failure.
5. Incumbent retaliation: ANDPAD or LIXIL ships a “free for solo” tier. Mitigation: build personal price-book moat fast; don’t compete on features, compete on the 5-minute LINE-native loop.

14. Structured verdict

Score:                  79/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       JA-fluent technical founder + a 元工務店 / 元リフォーム営業 advisor; 2nd engineer for backend + vendor catalog ops
Time to revenue:        8–12 weeks from soft launch (pre-sold pilot in parallel with build)
Capital to launch:      ¥4–6M (~$28–40K) — covers 2 engineers for 5 months, API + LINE costs, vendor catalog curation, IT subsidy registration paperwork
Top 3 assumptions to validate first:
  1. Solo contractors will send AI-generated 見積書 to customers with <90s of edits — 20-shop shadow pilot
  2. 5-min SLA is achievable at <¥80 per estimate — staged-input cost run
  3. Subsidy-led messaging lifts close rate ≥2x — split-arm cold outreach
Kill criteria:
  - Abandon if 4-week pilot shows >25% of generated PDFs require >5 minutes of contractor editing before send
  - Abandon if first 200 cold outreach attempts produce <2% trial-start rate even with subsidy framing
  - Abandon if year-1 cohort post-subsidy churn projects to >40% based on month 9-12 behavior
  - Abandon if a credible incumbent (ANDPAD, LIXIL, AnyONE) ships a free or sub-¥2K/mo solo tier within 6 months of MVP

15. Next step — 1-week validation sprint

• Day 1–2: Pull 30 contractors from リショップナビ + 外壁塗装パートナーズ directories. DM via their LINE official accounts with a 90-second demo video showing a fake-but-realistic 見積書 generated from 5 photos + a voice memo. Ask for a 15-min Zoom.
• Day 3–4: Run 8-10 of the booked Zooms. Show a Wizard-of-Oz demo (human-in-the-loop generates the PDF in 5 minutes while the contractor watches). Ask: would you pay ¥3,980/mo? Would you pay ¥800 effective with subsidy? What would make you switch from your current method (paper / Excel / nothing)?
• Day 5: Decide. Go if ≥4 of 10 contractors say “I’d pay today, send me the link” and ≥2 of those 4 explicitly mention subsidy as a closer. Pivot if signal is on photo-AI but not on LINE channel. Kill if <3 of 10 commit to pay or if everyone says “I already use [incumbent X] and it’s fine.”

The falsifiable outcome is paid-intent, not vibes. ¥1,000 deposit toward first month closes the loop.