← Sticker Wall AGENT · DETAIL

AI Invoice Classification Working Paper White-Label SaaS: Selling to Bookkeeping Firms for 46K RMB/Month

Workflow: Routinely pull bank statements and input invoices from bookkeeping firm clients' ERP systems every day. After AI perform

AGENT

Key Fields

FIELD STAMPS

IndustryFintech

RegionChina（中国大陆）

ScaleSME

ChannelOnline

🔧 Workflow

Routinely pull bank statements and input invoices from bookkeeping firm clients' ERP systems every day. After AI performs batch OCR recognition on invoice elements, it automatically classifies them by tax category, generates tax filing working papers with evidence links, pushes them to accountants for manual review, and writes them back to the accounting sets. Inputs are input invoices, bank statements, and ERP accounts; outputs are compliant working paper packages that can be directly imported into filing systems, with month-end automated summaries generating tax filing preparation sheets.

🛠 Setup Requirements

First, register on the Kingdee Cloud Star Open Platform and Baidu Intelligent Cloud OCR, write an invoice parsing service using Python to map invoice fields to working paper accounts; then set up an automated pipeline in n8n for scheduled statement pulling, OCR recognition, Feishu multi-dimensional table review, and working paper generation. Technical requirements include familiarity with API integration and basic front-end/back-end development. A usable version takes about 4 weeks, with the remaining 2 weeks spent using real invoices accompanied by accountants to calibrate error-prone account rules.

🧰 Toolchain

🔧 Baidu Intelligent Cloud VAT Invoice OCR
🔧 n8n Automated Workflow
🔧 Kingdee Cloud Star Open API
🔧 Feishu Multi-Dimensional Table
🔧 DeepSeek Large Model Fallback Rule Engine

💰 Revenue

① Small and medium bookkeeping firm white-label subscription (main revenue): Bookkeeping firms subscribe monthly based on the volume of monthly hosted invoices. 2,500 RMB/month × 15 bookkeeping clients = 37,500 RMB/month (self-calculated), accounting for about 82% of the disclosed monthly revenue of 46,000 RMB (this proportion is derived by estimation based on case materials, without independent verification, as of 2026). Client-side efficiency reference shows a 60% reduction in tax filing time (disclosed by the company itself), with the proportion of total revenue undisclosed; ② Annual upfront payment: The same batch of bookkeeping firms pays once annually. 2,500 RMB/month × 12 months = 30,000 RMB/client/year (estimated), with the first batch of 15 clients generating approximately 450,000 RMB upfront (estimated). This is subscription-sourced upfront cash flow, and its share is undisclosed (case perspective, lacking independent review); ③ Usage-based pricing markup: Hosted invoice volume exceeding the package is billed per piece. The exact price per piece and the excess threshold are not specified. Cost-side reference for invoice OCR is about 0.5 RMB/piece (cost item taken from self-description), and its proportion of revenue is unknown; ④ Opportunity to be validated—providing private deployment and working paper field library customization for small bookkeeping firms with 20 to 40 employees, charging a one-off service fee per project. Project quotes are not published, and their market share is also unknown.

💸 Cost

Invoice OCR is billed at approximately 0.5 RMB per piece, totaling about 10,000 RMB for 20,000 pieces per month; cloud servers and notification SMS cost about 500 RMB/month; channel partnership commission is about 10%.

⏱ Time Investment

Daily investment of 2 hours monitoring abnormal invoices and the review queue, with 1 concentrated day at month-end reconciling working papers with accountants; total weekly commitment is about 12 to 14 hours.

🚀 Getting Started

Step 1: Find a small bookkeeping firm with 20 to 40 employees for a pilot test. Take 500 real input invoices and corresponding tax filing receipts. Manually label the four columns of fields: year-month, tax category, tax rate, and deduction status to calculate the current manual time baseline. After creating a demonstrable automated classification working paper Demo, negotiate a subscription partnership of 2,000 to 3,000 RMB per month, lock in 3 trial clients to run for 2 months, and then scale replication.

🔑 Keys to Success

✅ Classification accuracy must remain stably above 97%, with major anomalies automatically routed to manual review with audit trails
✅ Track VAT policy changes monthly and continuously maintain the working paper field library
✅ Subscription billing based on the number of licenses hosted by the bookkeeping firm, with revenue growing alongside client scale
✅ Ensure good API compatibility with mainstream ERP systems like Kingdee and Chanjet to lower the client onboarding barrier
✅ Establish an invoice authenticity verification pipeline, integrating with the State Taxation Administration's invoice inspection API to ensure input compliance

⚠️ 风险

⚠️ Errors in tax filing working papers may bring tax legal risks; manual review and confirmation must be used as the delivery gate, and contracts must explicitly define the bookkeeping firm as the primary responsible entity for filing
⚠️ OCR recognition accuracy is affected by invoice clarity; blurry invoices may lead to classification errors, requiring a manual review queue and confidence thresholds
⚠️ Client data security and compliance risks for bookkeeping firms, requiring Level 3 Equal Protection certification (MLPS Level 3) and signed data confidentiality agreements
⚠️ API policy adjustments by platforms like Kingdee and Chanjet may cause interface instability, requiring backup data sources and multi-platform adaptation schemes

📌 Real Cases

📌 Kingdee AI Star Intelligent Tax Calculation Officer Case: Enterprise tax filing efficiency improved by 60%, month-end tax processing significantly simplified. This capability can now be reused by individual SaaS as a white-label channel; Chanjet Yi Daizhang's 2026 refreshed release simultaneously confirms that the bookkeeping industry has entered an AI dual-drive turning point.
📌 Glodon released Agentic 2.0 to reconstruct the 100-billion-RMB financial and tax service system, automatically processing invoices and reimbursement workflows through AI Agents, having completed pilot integration for multiple bookkeeping firms.
📌 nexu.io's AI bookkeeping agency solution: Automatically connects to enterprise ERP systems, achieving invoice OCR recognition, automatic account classification, and tax filing working paper generation, with processing efficiency several times higher than manual methods.
📌 Glodon (Simuw) Smart Accounting Tax Filing Agent: Cross-system data collection and automated filing, having served hundreds of micro and small enterprises with filing accuracy remaining stably above 98%.

Sources