What’s the difference between a FAQ bot and a RAG bot?

FAQ bots rely on static Q&A pairs; RAG queries updatable sources (documents, knowledge base) and returns answers with citations.

When is human handoff necessary?

For high‑risk cases, low confidence, or processes requiring verification; the transfer keeps context and history.

Deflection rate, CSAT/NPS, time to resolution, answer precision, escalation rate and cost per contact.

Knowledge with citations, intents & workflows, channel integration and human handoff with Azure AI.

AI for Business Cluster · Back to pillar · Language · Speech

Index docs/KB and answer with citations. Continuous updates and source control.

Intent detection, slot‑filling, orchestration and automations (tickets, orders, returns).

Web, mobile, WhatsApp, Teams, voice/IVR with Speech‑to‑Text and Text‑to‑Speech.

Escalation with transcript, confidence and context; CRM/ITSM integration.

Intent coverage, resolution rate, satisfaction, cost per contact and feedback loops.

RAI policies, sensitive‑category blocks, PII redaction, confidence thresholds and fallbacks.

User → Channel → Orchestrator → (RAG/Intent) → Action/Answer → Handoff → Feedback

Automate retraining on feedback and new knowledge.

Define evaluation rubrics, use user feedback and A/B testing.

Length limits, content filters, timeouts, retry/backoff and observability.

Concise answers, quick‑reply buttons, suggestions and consented memories.

Which KPIs to start tracking?

Deflection, CSAT, first contact resolution, escalation rate/time, cost per contact, answer quality.

How to handle privacy?

PII redaction, data classification, minimal retention, audit and end‑to‑end encryption.

How to handle traffic spikes?

Autoscaling, caching, async queues and rate limiting on channels.