Chat / Voice / Bots

💼 Business Context

• Conversational channels are now a primary support and sales surface — measuring them means measuring deflection, intent, and satisfaction, not just call volume
• Powers automated escalation and agent-assist, reducing average handle time
• Owned by Customer Support / Conversational AI teams

🔌 Technical Overview

Conversation platforms (Azure Bot Service, Intercom, custom bots, IVR-to-text) call a small .NET Core Azure Function acting as a webhook receiver, which authenticates the payload and forwards it through the shared Cxos.Ingestion.Client contracts into the Ingestion API. Intent classification runs either at the platform (Azure Bot Service's built-in NLU) or downstream in the Stream Worker via a call to the AI & Insights API, backed by Azure OpenAI Service. Transcripts are tagged for redaction before long-term storage.

💾 Sample Event Payload

{
  "event": "intent_detected",
  "event_id": "b3e1a2f0-6a9b-4b7f-9d2a-1a5c9e6b3f01",
  "timestamp": "2026-08-01T09:41:02Z",
  "user_id": "cust_004821",
  "context": {
    "channel": "chat",
    "conversation_id": "conv_9931",
    "platform": "web_widget"
  },
  "properties": {
    "intent": "return_request",
    "confidence": 0.93,
    "order_id": "ord_55123"
  }
}

🔗 Integration Points

• Azure Function (.NET Core isolated worker) as the platform webhook receiver
• Cxos.Ingestion.Client NuGet package for the outbound call into the Ingestion API
• Azure OpenAI Service — intent/sentiment classification, AI Copilot
• Activation API + Azure Service Bus — agent handoff and journey triggers

⚠️ Non-Functional Considerations

• Scale: bursty around support incidents/outages — Azure Functions consumption plan absorbs spikes without pre-provisioned capacity
• Latency: handoff/escalation events route through a dedicated high-priority Service Bus topic
• Reliability: conversation continuity survives a dropped connection via the platform's own session ID
• Security/Privacy: transcripts are redacted (PII/payment/health details) by an Azure Function before reaching the Data Lakehouse

🎯 Enterprise Example

A telecom provider's Azure Bot Service and IVR both forward transcripts through a shared .NET Core webhook receiver into the Ingestion API. The AI & Insights API, using Azure OpenAI Service, flags a spike in "billing_dispute" intents tied to a specific promotion within minutes instead of next week's report, and a temporary FAQ deflection flow is activated through the Activation API — cutting live-agent volume for that issue by 30%.