MarketRadar
← All ideas IDEA-036 6.7 japan api-as-a-product developed

POST /jp/invoice/extract → JSON with every インボイス制度 field native: 適格 number, split 消費税, 振込先, 源泉 — pay per call, integrates straight into freee/MF/弥生.

Problem

Japanese accountants run a 5-step manual workflow on every PDF invoice (email→Excel→freee CSV→batch import→reconciliation) because global OCR APIs (Mindee, Rossum, Textract) ignore JP-specific fields: 適格請求書 registration number, split 消費税 8%/10%/non-taxable, 振込先 bank routing, 源泉 withholding. Companies handling 200 invoices/month spend 10+ hours just on journal entry.

Audience

JP mid-market 中小企業 (100-500 invoices/mo, ¥30-100K/mo WTP), 会計事務所 (bookkeeping bureaus serving 50+ clients each), and B2B SaaS that need invoice processing as a feature on freee/MFクラウド/弥生.

Reasoning

Source signals

  • **SIG-20260425-japan-api-as-a-product-jp-invoice-pdf-extraction-api** — "5-step manual work, 10+ hours/month for 200 invoices, インボイス制度 fields not handled by global APIs."
  • **SIG-20260425-japan-api-as-a-product-freelance-automation-api-recipe** — "Same pain at solo-freelancer scale; DIY Python automation."
  • **SIG-20260425-japan-api-as-a-product-monk-ai-receivables-api** — "Receivables/payables reconciliation pain, freee API integration referenced."
  • **SIG-20260425-japan-api-as-a-product-jp-handwriting-jis-normalize-api** — "Global OCR fails on JP text/document specifics; engineers DIY with gpt-4o-mini."

Reasoning

5+ independent sources document the same pain over 18+ months, with a hard regulatory mandate (インボイス制度, October 2023 onward) ensuring it grows rather than fades. The build maps exactly onto our stack: Claude Sonnet 4.6 vision API + FastAPI + Postgres + R2 + Stripe metered billing, all on Railway + Cloudflare Pages, achievable in ~10 days. The moat is JP-specific schema fidelity + write-back integrations to freee/MF/弥生 that incumbents can't replicate quickly because it's outside their core competence. Two parallel monetisation models (pay-per-call for devs, subscription tiers for accounting firms) reduce single-channel risk.

Quick competitive read

  • **AIパンチャー** — JP-local but per-template, not per-call; not developer-targeted.
  • **Mindee / Rossum / AWS Textract** — global, weak on インボイス制度 fields.
  • **freee/MFクラウド built-in OCR** — quota-limited, our actual ICP is their power-users.
  • **AI Inside / DX Suite** — JP enterprise, ¥100K+/mo, heavy sales cycle, not developer-friendly.
  • No direct per-call developer-API competitor for JP invoices: **white space**, `competitive_pressure: 6-8`.

Initial pricing hypothesis

Pay-per-call $0.20/extract (Free 100/mo), Pro $49/mo (500 calls + webhooks), Studio ¥10K/mo for bookkeeping bureaus (3K calls + freee write-back), Enterprise ¥30K+/mo (unlimited + SLA + JPY/インボイス invoice).

Distribution hypothesis

Qiita posts from accounting-engineer founders ("how we killed 10 hours/mo of journal entry"), freee/MF App Marketplace listing as "invoice AI plugin", direct cold-outreach to ~100 会計事務所, Note.com/YouTube Japanese video demos.

What we ship

**Core features (MVP)**

  • Per-call extraction API: `POST /v1/jp/invoice/extract` accepts PDF/image, returns JSON with every インボイス制度 field — 適格請求書 13-digit T-number (validated against 国税庁 公表サイト), 消費税 split (8% reduced / 10% standard / non-taxable), 振込先 (bank, branch, account type, number), 源泉 withholding for 報酬, supplier+customer details, line items.
  • Free GUI demo + uploader playground on the landing page — drag-and-drop PDF, live JSON output side-by-side; doubles as the strongest pre-sales tool for the non-developer 会計事務所 audience.
  • インボイス制度 transitional credit countdown surfaced in the dashboard ("80% input-tax-credit period ends Sep 2026, 50% from Oct 2026, 0% from Oct 2029") — converts regulatory urgency into product CTA on every invoice missing a T-number.
  • Premium write-back into freee / MFクラウド / 弥生 — after extraction, API auto-creates the journal entry via freee API with the customer's 勘定科目 mapping, so the 会計事務所 sees the nota fiscal in their ERP with zero copy-paste.
  • JPY enterprise billing with 適格請求書 number — dashboard toggle for JPY mode at ¥10K/¥30K/¥100K tiers, PDF carries T-number + 消費税 10% breakdown, freee/MF-export for the customer's contador.

**Primary user flow**

1. Developer onboarding — GitHub OAuth → create project → API key → curl example with sample PDF in the playground (≤90 s).

2. First extract call — `POST /v1/jp/invoice/extract` with a PDF; <5 s response, JSON with every T-number / 消費税 / 振込 / 源泉 field plus per-field confidence score.

3. Webhook delivery (optional) — `POST /v1/webhooks` with URL → every extraction fires a HMAC-signed payload into the customer's system.

4. Billing — pay-per-call $0.20/extract (Free 100/mo) → Pro $49/mo (500 calls + webhooks) → Studio ¥10K/mo for 会計事務所 (3K calls + freee/MF write-back) → Enterprise ¥30K+/mo (unlimited + SLA + JPY/適格請求書 invoicing).

**What it looks like**

  • Landing page (developer-friendly): hero "Extract every インボイス制度 field in one API call — T-number, 消費税 split, 振込先, 源泉", live drag-and-drop demo with a real PDF and JSON in the right pane, transitional-credit countdown badge ("50% credit haircut in N days"), price $0.20/extract, side-by-side comparison vs Mindee (no T-number support) and AIパンチャー (per-template, not per-call).
  • API docs page: Stripe-style three-pane EN/JP, "Try it" runs against the user's real key, curl/Python/Node samples side-by-side, dedicated "インボイス制度 schema reference" mapping all 13 mandatory fields.
  • Dashboard: extractions/day with success/fail breakdown, confidence-score histogram per field, webhook delivery log, freee/MF write-back status, API key CRUD, Stripe billing portal, インボイス制度 transitional-credit countdown widget.
  • JPY billing screen: PDF preview with 適格請求書 number (T1234567890123), 消費税 (10%) breakdown, 振込先 routing, invoice history, freee/MFクラウド/弥生 export for the customer's contador.

**MVP build plan: 10 days**

  • Day 1-2: FastAPI + Postgres + R2 + Stripe metered billing, GitHub OAuth, scaffolded dashboard.
  • Day 3-4: Claude Sonnet 4.6 vision prompt for インボイス制度 schema (T-number, 消費税 split, 振込先, 源泉); GPT-4o-mini fallback for cost tier.
  • Day 5: T-number validation via 国税庁 公表サイト with Postgres cache; per-field confidence scoring.
  • Day 6: GUI uploader playground on the landing page (Cloudflare Pages), live JSON output renderer.
  • Day 7: Webhook delivery (Resend), HMAC signing, retry queue.
  • Day 8: freee API journal-entry write-back (premium tier), MFクラウド adapter.
  • Day 9: JPY invoice generator with 適格請求書 number, transitional-credit countdown widget.
  • Day 10: API docs polish, soft launch via Qiita post + 10 cold-outreach 会計事務所 pilots.