When should the AI hand off to a human?
Set keyword triggers, confidence thresholds, and turn-count limits so your AI agent hands off to a human agent with a full transcript and intent summary attached.
The agent is great at repetitive questions. It's deliberately cautious about anything it is not confident about — and in those cases, a human reply is a much better customer experience than a wrong automated one.
Default handoff triggers
- Customer types keywords like 'agent', 'manager', 'human', 'complaint', or any of your custom escalation words.
- The agent's confidence score drops below the threshold you set (default 0.7).
- The conversation has exceeded N turns without resolution (default 6).
- Any mention of refunds, cancellations, or legal action.
- The customer's sentiment score turns negative.
When handoff fires, the agent hands the agent the full transcript, the customer's current intent in one line, and the next-best action it was about to suggest — so the human never starts from zero.
Related articles in AI WhatsApp Agents
Browse other help categories
Getting started
Create your SMSLocal account, add INR credits to your wallet, complete DLT registration, and send your first compliant SMS, WhatsApp, or OTP in under fifteen minutes.
SMS campaigns
SMS content templates, DLT routing, sender ID setup, delivery report statuses, bulk campaign scheduling, throttle controls, and automatic retry logic for India.
WhatsApp Business API
Complete guide to WhatsApp Business API — onboarding, Meta template approval, chatbot flow setup, shared team inbox, quality rating management, and conversation analytics.
OTP & Transactional
Priority routing to the fastest carrier, automatic retry and failover logic, idempotency keys to prevent double-sends, and WhatsApp OTP fallback when SMS is slow or fails.
Did this article help?
If something isn't clear or the steps don't match what you're seeing, tell us and we'll fix the doc the same day.