Email-forward → AI line-item extraction → auto-push to QuickBooks/Xero — vertical-tuned for one industry's invoice quirks, $39/mo flat.

What we ship

**Core features (MVP)**

• Email-forward setup wizard: a unique `invoice@yourname.parserbox` address; wizard walks through Gmail/Outlook auto-forward setup (and the @qbodocs.com verification workaround).
• Vertical Chart-of-Accounts mapping table: ad-agencies ship with a ready table ("media buy → 5040 Media Costs", "ad-spend rebate → 4900 Other Income"); clients can override rows.
• AI line-item extraction: Claude vision API parses the PDF — vendor, total, line-items, tax. Works on non-standard layouts that defeat generic OCR.
• Line-item review screen: split-view with PDF on the left, extracted fields on the right; inline editing and an "Approve & push to QuickBooks" button.
• Auto-push to QuickBooks/Xero: creates a Bill via read-write `bills` scope, never touches GL.

**Primary user flow**

1. SMB owner / bookkeeper finds us via QuickBooks App Store listing or LinkedIn outreach.

2. Setup wizard in 5 minutes: connects QuickBooks Online OAuth, picks a vertical (ad-agency), configures Gmail forward to `invoice@`.

3. Forwards first 5 vendor invoices; within 30 seconds they appear in the queue with extracted line-items, hits Approve on the review screen.

4. By day 3 manual retyping is gone; on day 14 converts from trial to $39/mo Solo.

**What it looks like**

• **Landing page**: hero "Forward your invoices. We do the typing." Bullets for ad-agencies (media-buy COA ready), restaurant-supply (COGS ready). Screenshot of the review screen. CTA: "Connect QuickBooks — try free for 14 days".
• **Main app screen**: 3-tab layout — (1) Inbox: queue of invoices with status chips (new / needs review / approved / pushed), (2) Review: split PDF + extracted fields with inline editing + COA mapping table, (3) Settings: forward address, OAuth, COA table, 90-day retention toggle.
• **Output format**: every processed invoice becomes a QuickBooks Bill with the correct GL code + attached PDF; weekly email summary "12 invoices processed, $4,302 total, 2 hours saved".

**MVP build plan (12 days)**

• Day 1-2: FastAPI + Postgres + Celery (Redis) on Railway, JWT, Stripe Subscriptions, Cloudflare R2 for PDFs.
• Day 3-4: Mailgun/Postmark inbound webhook, parse forwarded email, save PDF to R2.
• Day 5-6: Claude Sonnet 4.6 vision pipeline for line-item extraction, structured JSON output, retry on partial fails.
• Day 7: Vertical-tuned prompt + ad-agency COA mapping, fallback to default COA.
• Day 8-9: React + Tailwind UI — Inbox, Review (split-view PDF + fields), Settings, COA-table editor.
• Day 10: QuickBooks Online OAuth + Bill creation API; Xero OAuth + ApInvoice API.
• Day 11: Email-forward setup wizard, landing page on Cloudflare Pages, DPA template.
• Day 12: ProductHunt + QuickBooks App Store assets, end-to-end smoke test, ship.

Source signals

• **SIG-20260425-us-micro-saas-utility-vendor-invoice-ocr-sweet-spot**: "Spent 6 hours manually entering supplier invoices... I still end up manually entering maybe 20% regardless of what tool I use" — RECURRING, 3+ 2026 launches
• **SIG-20260425-us-micro-saas-utility-spreadsheet-replacement-niche** (adjacent): "I just use a spreadsheet for X" — bookkeepers tracking invoices in spreadsheets is a textbook spreadsheet-replacement target

Reasoning

The vendor-invoice gap has documented 6+ hr/week pain, three 2026 launches (Parsli, Expensent, Syntora) targeting the same wedge, and a structural QuickBooks flaw (@qbodocs.com requires manual human forward). What's missing from all three competitors: vertical-tuned prompts. Generic AP-automation produces a flat extraction; an ad-agency bookkeeper needs media-buy line-items mapped to specific Chart-of-Accounts codes; a restaurant owner needs ingredient categories mapped to COGS sub-accounts. By picking ONE vertical first (ad-agencies has the highest WTP / lowest churn profile in our research), we get a defensible niche before generalizing. Stack fits Railway + CF Pages cleanly: Mailgun inbound email → Celery → Claude vision → Postgres → QuickBooks/Xero OAuth. Build 12 days. Recurring revenue is structurally durable: invoices arrive monthly, churn is low while the business operates.

Quick competitive read

• **Dext (Receipt Bank)** — $37.50/250 docs, 14-yr incumbent, horizontal, generic COA
• **Hubdoc (Xero)** — free-with-Xero / $12 standalone, 3.3⭐, stagnant, Xero-only-friendly
• **Parsli** — 2024 AI-first OCR, no native QBO/Xero (Zapier-only), per-page pricing
• **Ramp Bill Pay** — free for Ramp card customers, generic, gates SMB without cards
• **QuickBooks Online @qbodocs.com gateway** — broken for auto-forward, manual-only

Market density: **medium-high (5+ players)** but each has a structural gap: Dext stays horizontal, Hubdoc stagnates, Parsli lacks native QBO integration, Ramp requires its card program. **White-space confirmed**: vertical-specific Chart-of-Accounts mapping (ad-agencies → media-buy GL codes, restaurant supply chains → COGS sub-accounts) packaged as turn-key QBO/Xero plugin at $39 flat. Start with ad-agencies (highest WTP / lowest churn).

Initial pricing hypothesis

$39/mo Solo (up to 50 invoices/mo, 1 QuickBooks account) → $79/mo Pro (up to 250 invoices, multi-account, custom approval workflow) → $149/mo Bookkeeper (multi-client workspace, white-label PDFs). Anchored on developed-tier WTP and positioned between manual ($0) and enterprise AP ($1000+).

Distribution hypothesis

Three channels:

1. **Cold-email outreach to bookkeepers** on LinkedIn — each bookkeeper serves 5-15 SMB clients, converting one = 5-15 seats

2. **Content-SEO** — "QuickBooks vendor invoice automation for [industry]" long-tail

3. **QuickBooks App Store + Xero App Store** listings — high-intent discovery, low CAC