WhatsApp chatbot with AI and human handoff to a real secretary (Make + WhatsApp)

Beginner Questions
, , ,
1

:bullseye: What is your goal?

My goal is to build a hybrid WhatsApp chatbot using Make and WhatsApp that works like a real secretary:

First, an AI assistant answers general and frequently asked questions.

When the AI needs to collect personal data from the client (or when the client asks to talk to a human), the AI should stop responding.

At that point, a real human secretary should take over the WhatsApp conversation.

After 24 hours of no messages, if the client writes again, the AI should automatically resume and handle the conversation again.

:thinking: What is the problem?

I have successfully implemented AI responses using GPT / Make AI Agent, but I haven’t been able to properly manage the handoff between AI and a human secretary.

Specifically:

I don’t know the best way to disable the AI for a specific chat.

I’m not sure how to notify my secretary and allow her to continue the same WhatsApp conversation.

I also need a reliable way to reactivate the AI after 24 hours of inactivity if the client messages again.

:test_tube: What have you tried so far?

Connected WhatsApp to Make and set up message triggers.

Implemented AI replies using GPT / Make AI Agent.

Tried using filters and basic flags, but I haven’t found a clean way to:

Store the conversation state (AI vs Human)

Pause AI responses during human takeover

Automatically switch back to AI after 24 hours

3

If you are open to use a third-party tool, look at https://www.handoff.chat/.

4

I’m looking at a similar WhatsApp AI → human handoff pattern from the builder side.

One thing I’m trying to separate: when this breaks in production, is the reusable part mainly the Make scenario/routing logic, or the handoff artifact the secretary receives after escalation?

For example: summary, intent, missing info, risk flags, source snippets, draft reply, and the next human action.

If you were adapting this for another client, would you rather start from a reusable Make scenario, a handoff field/schema checklist, a reusable prompt/agent instruction for the handoff, or a short review of the current flow?

5

I would model this as a small state machine, not only as filters.

For each WhatsApp contact, store a row like:

  • contact_id
  • mode: ai / human
  • mode_until
  • handoff_reason
  • last_summary
  • missing_fields
  • next_human_action
  • owner_or_secretary

Then the Make scenario checks this row before every AI response. If mode=human, the AI exits silently and only the secretary path runs. If the customer is inactive for 24h, the mode can expire back to ai.

The practical issue is the handoff packet. The secretary should not receive just a transcript; she needs a short brief with what the client wants, what data is missing, the risk/urgency, and the exact next action.

I am testing whether this should be packaged as an importable Make/n8n workflow plus handoff schema and setup guide. For your case, would you pay for a ready-to-adapt template, or would you only want done-for-you setup?