How to Automate Accounts Payable with n8n and AI (German Mittelstand Guide)

Definition

Accounts payable automation for the German Mittelstand combines invoice intake (email, portal, E-Rechnung), AI-powered data extraction, approval routing, and ERP/DATEV export, orchestrated by workflow tools like n8n without requiring SAP or enterprise AP platforms.

Why n8n for Mittelstand AP

Enterprise AP solutions (Basware, Tipalti, Yokoy) cost €50K+ annually and assume SAP/Oracle backends. German Mittelstand companies: 50 to 500 employees, €10M–€100M revenue: need automation that connects to DATEV, Lexware, or SevDesk without six-figure contracts.

n8n provides:

• Self-hosted data control (GDPR, GoBD)
• Visual workflow builder finance teams can modify
• 400+ integrations including email, Slack, HTTP, databases
• AI node support for Claude/GPT extraction
• One-time setup cost, low ongoing fees

Architecture Overview

Invoice Intake → Format Detection → Extraction → Validation → Approval → Export
     ↓                  ↓               ↓            ↓           ↓          ↓
  Email/IMAP      XRechnung/ZUGFeRD   AI/OCR     Rules Engine  Slack    DATEV CSV
  Supplier Portal    PDF scan       Claude      PO Match     Email    Lexware API

Step 1: Invoice Intake Workflow

Create an n8n workflow triggered by email (IMAP) or webhook:

{
  "nodes": [
    {
      "name": "Email Trigger",
      "type": "n8n-nodes-base.emailReadImap",
      "parameters": {
        "mailbox": "rechnungen@company.de",
        "downloadAttachments": true
      }
    },
    {
      "name": "Detect Format",
      "type": "n8n-nodes-base.code",
      "parameters": {
        "jsCode": "const items = $input.all();\nreturn items.map(item => {\n  const filename = item.binary?.attachment_0?.fileName || '';\n  const mime = item.binary?.attachment_0?.mimeType || '';\n  let format = 'unknown';\n  if (mime === 'application/xml' || filename.endsWith('.xml')) format = 'xrechnung';\n  else if (mime === 'application/pdf') format = 'pdf';\n  return { json: { ...item.json, format, filename } };\n});"
      }
    }
  ]
}

Route by format: XML → direct parsing, PDF → AI extraction, ZUGFeRD → embedded XML extraction first, fallback to AI.

Step 2: AI Invoice Extraction

For PDF invoices without structured data, use Claude with structured output:

// n8n Code node - call Claude API
const pdfBase64 = items[0].binary.attachment_0.data;

const response = await fetch('https://api.anthropic.com/v1/messages', {
  method: 'POST',
  headers: {
    'x-api-key': $env.ANTHROPIC_API_KEY,
    'anthropic-version': '2023-06-01',
    'content-type': 'application/json'
  },
  body: JSON.stringify({
    model: 'claude-sonnet-4-20250514',
    max_tokens: 2000,
    messages: [{
      role: 'user',
      content: [
        {
          type: 'document',
          source: { type: 'base64', media_type: 'application/pdf', data: pdfBase64 }
        },
        {
          type: 'text',
          text: `Extract invoice data as JSON with fields:
            rechnungsnummer, rechnungsdatum, faelligkeitsdatum,
            lieferant_name, lieferant_ustid, nettobetrag, ust_betrag,
            bruttobetrag, waehrung, positionen (array of beschreibung, menge, einzelpreis, steuersatz).
            Return ONLY valid JSON.`
        }
      ]
    }]
  })
});

const extracted = JSON.parse(response.content[0].text);
return [{ json: extracted }];

Set confidence thresholds: auto-approve extractions above 95% field match against vendor master data. Route lower confidence to human review.

Step 3: Validation Rules

Before approval, validate extracted data:

// Validation code node
const invoice = $input.first().json;
const vendor = await $('PostgreSQL').execute({
  query: 'SELECT * FROM vendors WHERE ust_id = $1',
  params: [invoice.lieferant_ustid]
});

const errors = [];
if (!vendor) errors.push('Unknown vendor USt-IdNr.');
if (vendor && vendor.default_iban !== invoice.lieferant_iban) {
  errors.push('IBAN changed from vendor master');
}

const duplicate = await $('PostgreSQL').execute({
  query: 'SELECT id FROM invoices WHERE number = $1 AND vendor_id = $2',
  params: [invoice.rechnungsnummer, vendor?.id]
});
if (duplicate.length) errors.push('Duplicate invoice');

return [{ json: { ...invoice, validation_errors: errors, status: errors.length ? 'review' : 'approved' } }];

Step 4: Approval Routing

Use n8n Switch node for approval matrix:

• < €500: Auto-approve if validation passes
• €500–€5,000: Slack approval to department head
• €5,000–€25,000: Email + Slack to finance manager
• > €25,000: Sequential approval (FM → CFO)

{
  "name": "Approval Router",
  "type": "n8n-nodes-base.switch",
  "parameters": {
    "rules": {
      "rules": [
        { "value": "={{ $json.bruttobetrag }}", "operation": "smallerEqual", "output": 0, "value2": 500 },
        { "value": "={{ $json.bruttobetrag }}", "operation": "smallerEqual", "output": 1, "value2": 5000 },
        { "value": "={{ $json.bruttobetrag }}", "operation": "smallerEqual", "output": 2, "value2": 25000 }
      ]
    }
  }
}

Wait for approval via n8n Wait node with webhook callback from Slack interactive buttons.

Step 5: DATEV Export

Generate DATEV-compatible CSV (EXTF Buchungsstapel):

const invoice = $input.first().json;
const datevRow = [
  '11000',                              // Umsatz
  'H',                                  // Soll/Haben
  'EUR',                                // Währung
  '',                                   // Kurs
  '',                                   // Basisumsatz
  '',                                   // WKZ Basis
  invoice.nettobetrag.toFixed(2).replace('.', ','),  // Netto
  '8400',                               // Gegenkonto (Aufwand)
  '0000',                               // BU-Schlüssel
  formatDate(invoice.rechnungsdatum),   // Belegdatum
  invoice.rechnungsnummer,              // Belegnummer
  '',                                   // Belegtext
  '1',                                  // Postensperre
  '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', ''
].join(';');

await writeFile(`/exports/datev_${Date.now()}.csv`, datevRow);

Upload to DATEV Unternehmen online via DATEVconnect API or manual import: depending on your DATEV package.

Step 6: E-Rechnung Processing

For XRechnung/ZUGFeRD, skip AI extraction when structured XML is available:

# External Python service called via n8n HTTP node
from lxml import etree

def parse_xrechnung(xml_bytes):
    ns = {'cbc': 'urn:oasis:names:specification:ubl:schema:xsd:CommonBasicComponents-2',
          'cac': 'urn:oasis:names:specification:ubl:schema:xsd:CommonAggregateComponents-2'}
    root = etree.fromstring(xml_bytes)
    return {
        'rechnungsnummer': root.find('.//cbc:ID', ns).text,
        'rechnungsdatum': root.find('.//cbc:IssueDate', ns).text,
        'lieferant_name': root.find('.//cac:AccountingSupplierParty/cac:Party/cac:PartyName/cbc:Name', ns).text,
        'lieferant_ustid': root.find('.//cac:AccountingSupplierParty/cac:Party/cac:PartyTaxScheme/cbc:CompanyID', ns).text,
        'bruttobetrag': float(root.find('.//cbc:TaxInclusiveAmount', ns).text),
        'nettobetrag': float(root.find('.//cbc:TaxExclusiveAmount', ns).text),
    }

Structured extraction is faster, cheaper, and more accurate than AI: always prefer XML when available.

Monitoring and Metrics

Track in n8n or connected Grafana:

Common Pitfalls

• Running n8n cloud without self-hosting for sensitive invoice data (GDPR concern)
• No vendor master data maintenance: AI extraction cannot validate unknown vendors
• Skipping IBAN change detection: primary fraud vector in AP
• Manual DATEV CSV without automated upload: creates second bottleneck
• Over-automating first invoice from new vendors: always require human review

Business Outcomes

Mittelstand companies implementing this stack process invoices in under 5 minutes versus 15–20 minutes manually, reduce AP headcount needs by 1–2 FTE, achieve 99%+ duplicate detection, and maintain GoBD-compliant audit trails through n8n execution logs.

Related: E-Rechnung 2025 Guide · E-Rechnung integration