I built a Claude Code skill that processes ecommerce accounting documents and generates DATEV-ready bookkeeping outputs automatically.

Instead of manually categorizing thousands of invoices, the AI:

• reads invoices
• categorizes expenses
• applies German VAT rules
• maps entries to DATEV accounts
• generates structured bookkeeping outputs

The system processed thousands of documents and produced structured accounting files in minutes.

---

The Problem

Ecommerce accounting is extremely fragmented.

A typical German ecommerce stack includes:

• Amazon / Etsy / Shopify
• PayPal / Stripe / payment processors
• supplier invoices
• VAT reports
• marketplace settlement reports
• EU OSS reporting

Even if documents are organized by month, processing them manually is slow and error-prone.

The biggest bottleneck is always the same:

Categorizing supplier invoices and preparing structured bookkeeping data for the Steuerberater.

---

The Idea

Instead of treating bookkeeping as manual work, I built an AI skill system that mirrors how ecommerce accounting actually works.

The skill is designed for:

• German ecommerce companies
• EU VAT / OSS workflows
• DATEV accounting
• Amazon / Shopify / Etsy sales
• payment processors like Stripe and PayPal

The goal:

Transform raw accounting documents into structured bookkeeping data automatically.

---

The Architecture

The system is built as a modular AI bookkeeping toolkit.

/ai-ecommerce-accounting /skills invoice_reader settlement_parser payment_reconciliation vat_engine datev_mapper anomaly_detector

Each module solves a specific accounting problem.

---

Invoice Reader

Reads supplier invoices and extracts:

• supplier
• invoice number
• VAT rate
• net / gross values
• currency

Then categorizes the expense and assigns a DATEV account.

Example:

Supplier: DHL Category: Shipping Account: 4920 VAT: 19%

---

Marketplace Parser

Handles Amazon / Etsy / Shopify settlement reports.

Breaks reports into structured components:

• product revenue
• commissions
• shipping charges
• refunds
• marketplace fees

---

Payment Reconciliation

Processes payment processor reports such as Stripe and PayPal.

Example logic:

Customer payment → revenue Processor fee → payment fees Net payout → bank reconciliation

---

VAT Engine

Applies EU and German VAT rules including:

• German VAT
• EU OSS
• reverse charge
• cross-border transactions

---

DATEV Mapper

Converts all structured accounting data into DATEV-compatible bookkeeping entries.

Example output:

Date: YYYY-MM-DD Account: 4920 Description: Shipping expense VAT: 19%

---

Dataset

The pipeline processed a full archive of ecommerce accounting documents including:

• thousands of supplier invoices
• marketplace settlement reports
• ecommerce platform exports
• payment processor reports
• tax and VAT documentation

---

Results

The system produced structured accounting outputs including:

• categorized expense reports
• marketplace revenue breakdown
• payment processor reconciliation
• VAT summaries
• anomaly reports
• DATEV-compatible bookkeeping exports

All outputs were packaged into a structured dataset for accounting review.

---

Quality Control

The AI system automatically flags uncertain items for manual review, such as:

• missing values
• non-EUR currencies
• unclear suppliers
• inconsistent VAT calculations

This allows the Steuerberater to focus only on edge cases.

---

Time Savings

Manual bookkeeping for datasets like this normally takes dozens of hours.

The AI pipeline converts raw documents into structured bookkeeping data in minutes.

The accountant only needs to review flagged items.

---

Lessons Learned

1. Invoice extraction is the easy part

The real complexity is:

• VAT logic
• marketplace settlements
• payment reconciliation

---

2. Ecommerce accounting requires domain knowledge

Generic AI systems fail because they do not understand:

• marketplace fees
• OSS VAT
• payment processor reports
• settlement structures

Adding domain rules dramatically improves accuracy.

---

3. AI works best as a preprocessing layer

The goal is not replacing the Steuerberater.

The goal is transforming:

unstructured documents → structured accounting data.

---

Conclusion

AI coding tools like Claude Code make it possible to build domain-specific automation systems very quickly.

Instead of generic assistants, the real opportunity lies in specialized AI skills built around real workflows.

In this case:

AI becomes a preprocessing engine for ecommerce accounting.