I built an AI Receptionist that answers incoming WhatsApp messages, runs simple triage with an AI, and routes tricky cases to a human for handling. It runs on n8n, uses a WhatsApp Business API provider for messaging, and keeps a short-lived conversation state so follow-ups make sense. This guide shows how I wired the pieces, how I decide when a human steps in, and the safeguards I use to keep interactions sensible.
Start by wiring WhatsApp to n8n. Pick a WhatsApp Business API provider that gives a webhook for incoming messages. In n8n create a webhook node that accepts messages and extracts the phone number, message id, timestamp and text. Add a simple session store, for example a Redis key per phone number with a 30-minute TTL. Use an HTTP request node to send the message text and recent context to your AI model. Keep prompts concise and include the last 2 turns of context only. I use these steps in the flow:
- Receive webhook and normalise payload into a standard JSON shape.
- Check session data in Redis and attach the last 1–2 messages.
- Call the AI via the HTTP request node with a prompt template and a max token limit.
- Evaluate the AI response and confidence field in a Switch node to choose an action.
Make sure the WhatsApp provider supports templates for outbound messages you must send before a user replies, and store template IDs in a credential vault, not in plaintext.
Human-in-the-loop is the safety valve. I pick a confidence threshold from the AI that triggers an escalation. For plain intents like “opening hours” or “directions”, accept auto-replies if confidence is above 0.7. If confidence falls below 0.6, or if the message contains sensitive markers such as “complaint”, “refund”, “legal” or numbers longer than eight digits, create a human task. That task is a compact JSON card with the conversation history, suggested reply, and an action set: approve, edit and send, or escalate externally. I deliver that card to a lightweight queue. I have used an n8n HTTP node to POST the card to a small web dashboard, and a Slack webhook for faster triage. When a person edits or approves the reply, n8n receives the approval webhook, updates the Redis session, sends the reply through the WhatsApp provider, and logs the resolution. For verification, include a status field on the session object and reject automated send attempts if status is “pending human”. That prevents the AI from double-sending while a human is deciding.
Tuning the AI workflows and keeping interactions smooth takes work. Keep prompts explicit about tone, length and disallowed content. Use short quick-reply options to channel user intent; these reduce free-text ambiguity and lift AI confidence. Log every message, AI output and action with a unique conversation id; capture timing so you can measure average human handover time. Audit a random 5–10% of escalations weekly to spot prompt failures or misrouted intents. For complex queries, send a short acknowledgement to the user while the human handles it, for example: “I’ve passed this to an advisor, they’ll reply soon.” That keeps the chat live and avoids repeat messages from the user. Make sure data retention and consent rules are followed in your region; store minimal PII and delete session keys after 30–90 days.
Practical checkpoints that helped me: restrict AI replies to 1–2 sentences for the receptionist role, flag numeric or legal-looking content for human review, and use quick-reply buttons for actions like “Book a meeting” or “Request a quote”. Track three metrics: automated resolution rate, average human response time, and percentage of escalations that required further escalation. Tweak the confidence thresholds based on those numbers rather than guessing. The setup runs lean in n8n, uses the WhatsApp webhook as the single ingress, and treats human reviewers as an on-demand safety net. That combination keeps conversations fast, reduces silly AI mistakes, and lets a human fix the odd edge case without breaking the flow.
