KI-Integration IT-Systeme: 20% schneller + €58k ROI [ERP/CRM 2025]

KI_Integration_Stack:
  # Layer 1: Bestehende Systeme
  Datenquellen:
    - ERP: SAP Business One (REST-API)
    - CRM: Salesforce (SOAP + REST)
    - DMS: SharePoint (Microsoft Graph API)
    - Shop: Shopify (Webhooks)
    - E-Mail: IMAP/SMTP
  
  # Layer 2: Integration-Middleware
  Event-Bus:
    Technology: Apache Kafka ODER RabbitMQ
    Purpose: Async Events (Order Created, Invoice Uploaded)
    Throughput: 1000 Events/Sek
  
  API-Gateway:
    Technology: Kong ODER Azure API Management
    Purpose: Auth, Rate-Limiting, Routing
    Endpoints: 12 definiert
  
  # Layer 3: KI-Services
  KI_Microservices:
    - Document-AI: OCR + Extraktion (Rechnungen, Verträge)
    - NLP-Classifier: E-Mail/Ticket-Routing
    - Forecast-Service: Demand/Churn-Prediction
    - RAG-Search: Semantic Search über ERP/CRM
    - Validation: Plausibilitätschecks (Beträge, Datumslogik)
  
  # Layer 4: Governance & Monitoring
  Observability:
    - Prometheus + Grafana (Metriken)
    - ELK-Stack (Logs: Elasticsearch, Logstash, Kibana)
    - Jaeger (Distributed Tracing)
  
  Security:
    - OAuth 2.0 / OIDC (Keycloak)
    - API-Keys verschlüsselt (Azure Key Vault)
    - PII-Redaction (automatisch in Logs)

Kosten (monatlich):
  - Kafka/RabbitMQ: €80 (Cloud) ODER €0 (Self-Hosted)
  - API-Gateway: €150 (Kong Konnect) ODER €0 (Self-Hosted)
  - KI-Services (Azure/AWS): €220
  - Monitoring: €50
  - **Total: €500/Mt = €6.000/Jahr**

# Event-Flow: E-Mail-Bestellung → SAP

1. E-Mail-Eingang:
   - Trigger: IMAP-Watcher (n8n / Zapier)
   - Aktion: E-Mail an Document-AI-Service senden

2. Document-AI-Service:
   - NLP-Extraktion: Kunde, Produkte, Mengen, Lieferdatum
   - Plausibilitätschecks: Kunde existiert? Produkte auf Lager?
   - Confidence-Score berechnen

3. Routing (API-Gateway):
   - IF Confidence >= 0.90:
       → Event "order.create.auto" an Kafka
   - ELSE:
       → Event "order.review.required" an CRM (Task für Sales)

4. SAP-Integration (Event-Consumer):
   - Liest "order.create.auto" aus Kafka
   - REST-API-Call an SAP: Auftrag anlegen
   - Response: Order-ID zurück

5. Bestätigungsmail:
   - Template: "Ihre Bestellung {{order_id}} ist eingegangen"
   - Versand: SMTP-Service

6. Monitoring:
   - Event geloggt (ELK)
   - Metrik "orders_automated" +1 (Prometheus)
   - Alert bei Fehler (PagerDuty)

# KI-Integration: E-Mail → Document-AI → SAP

from fastapi import FastAPI
from azure.ai.formrecognizer import DocumentAnalysisClient
import requests

app = FastAPI()

# Document-AI Service
doc_ai = DocumentAnalysisClient(endpoint=AZURE_ENDPOINT, credential=AZURE_KEY)

# SAP-API Connector
class SAPConnector:
    def __init__(self, base_url, auth):
        self.base_url = base_url
        self.auth = auth  # Basic Auth ODER OAuth
        
    def create_order(self, order_data):
        """
        Auftrag in SAP anlegen
        """
        response = requests.post(
            f"{self.base_url}/b1s/v1/Orders",
            json={
                "CardCode": order_data['customer_id'],
                "DocDueDate": order_data['delivery_date'],
                "DocumentLines": [
                    {
                        "ItemCode": item['product_id'],
                        "Quantity": item['quantity'],
                        "UnitPrice": item['price']
                    }
                    for item in order_data['items']
                ]
            },
            headers={"Authorization": self.auth}
        )
        
        return response.json()

sap = SAPConnector(base_url="https://sap.company.local:50000", auth="Bearer TOKEN")

@app.post("/webhook/email-order")
async def process_email_order(email_content: str):
    """
    Webhook-Endpoint für E-Mail-Bestellungen
    """
    # 1. Document-AI: E-Mail-Inhalt analysieren
    result = doc_ai.begin_analyze_document("prebuilt-invoice", email_content).result()
    
    # 2. Datenextraktion
    extracted = {
        'customer_id': result.fields.get('CustomerID').value,
        'delivery_date': result.fields.get('DeliveryDate').value,
        'items': [
            {
                'product_id': item.fields.get('ProductID').value,
                'quantity': item.fields.get('Quantity').value,
                'price': item.fields.get('UnitPrice').value
            }
            for item in result.fields.get('Items').value
        ]
    }
    
    # 3. Validation (Plausibilitätschecks)
    confidence = result.confidence
    
    if confidence >= 0.90:
        # Auto-Create in SAP
        sap_response = sap.create_order(extracted)
        
        return {
            'status': 'auto_created',
            'order_id': sap_response['DocNum'],
            'confidence': confidence
        }
    else:
        # Human-Review nötig
        # Ticket in CRM anlegen
        return {
            'status': 'review_required',
            'reason': f'Low confidence ({confidence:.0%})',
            'extracted_data': extracted
        }