Practical guide

How WhatsApp AI-to-human handoff works

Learn when AI should pause, what customers should be told, and how your team can follow up on a request to speak to a real person.

At a glance

An AI-to-human handoff occurs when a customer asks for a person or the automated assistant cannot reliably resolve the request. RecoverPay records a pending handoff and guides the AI to acknowledge the need for a human. Team members must review the request and respond; an acknowledgement is not a completed resolution.

Respect a direct request

A customer asking for the owner or a real person should receive a clear acknowledgement.

Avoid invented promises

An assistant should not claim the owner is available or provide a callback time that has not been confirmed.

Preserve context

The team can read the conversation before answering instead of asking the customer to start again.

How to put it into practice

  1. Recognise the request

    Requests for a person, uncertainty about an answer, or sensitive account-specific issues can need human review.

  2. Acknowledge briefly

    Explain that a team member is needed. If the business is closed, say so using its configured hours.

  3. Review the pending request

    A team member checks the lead and the relevant WhatsApp conversation.

  4. Reply as a person

    Address the question directly. A successful human reply clears the matching pending handoff for that line.

Questions and answers

Will it send a WhatsApp alert to the owner?

A customer asking for a human creates a workspace handoff and uses the app’s notification settings. Do not assume it sends a separate WhatsApp message, SMS, or phone call to the owner.

Should the AI say “not available” to every question?

No. It should answer questions supported by the business context and ask for human help when necessary. Only a request that needs a person should be treated as a handoff.

What should the owner configure?

Set the correct business identity and hours, review the AI context, keep the relevant notification preferences enabled, and agree on who checks pending requests.