"Schedule for the 15th."
| Challenge | Description | Mitigation | | :--- | :--- | :--- | | | Customers feel deceived if not told they speak to AI. | Mandatory upfront disclosure: "This is an AI assistant from [Company]." Follows EU AI Act requirements. | | Hallucination Risk | LLM may invent incorrect appointment times or rates. | Ground the LLM with strict retrieval-augmented generation (RAG) from a verified knowledge base. No factual generation allowed. | | Emotion Misread | AI may misinterpret sarcasm or distress. | Human-in-the-loop for high-sentiment calls; confidence threshold (e.g., <0.8 sentiment confidence → flag for review). | | Telephony Latency | >300ms delay destroys conversation flow. | Deploy edge inference (on telecom gateways) rather than cloud-only. Use ultra-low-codec ASR. | conversational ai dialer