ConversaLabs Help Center
en
Language
pt_BR Portuguese (Brazil) en English es Spanish

AI with human approval (human-in-the-loop)

Open in
View as Markdown Open in ChatGPT Open in Claude
Conversa Labs

Conversa Labs

Last updated on Jun 27, 2026

Overview

A cadence step can be composed by you, by the Maestro AI, or by a mix of both. The highlight is AI assisted (human-in-the-loop) mode: the AI drafts the step's message, but nothing is sent without a person approving it — and that person can still edit the draft before sending.

This gives you the speed and personalization of AI without giving up control and your brand's tone. And if the Maestro is unavailable, the cadence doesn't stall: it uses the step's static text as a fallback.

Prerequisites

  • The Follow-ups module enabled and a sequence with steps.
  • The Maestro enabled on the account (without it, AI modes fall back to static text).
  • A user responsible for reviewing and approving drafts (the notification recipient).
  • For best quality, a step with a clear instruction/intent for the AI.

Step by step

  1. In the step editor, choose the compose mode: static, AI assisted or AI full.
  2. For AI, describe the step's intent (what the message should achieve) and leave a static fallback text.
  3. Publish the sequence. When an AI assisted step is due, the AI creates a draft and the step stays pending, awaiting approval.
  4. The responsible user receives a notification of a pending draft.
  5. Review the draft, edit it if you want, and approve to send — or discard it.

Settings & options

The three compose modes

Mode Behavior
Static You write the text (with variables). It's sent as is.
AI assisted The AI generates a draft and the step waits for human approval before sending. The draft can be edited.
AI full The AI composes and sends automatically, with no approval wait.

Human approval (HITL)

In AI assisted mode, the step creates a draft and pauses at that point. The reviewer can:

  • Approve the draft as is;
  • Edit the text and then approve (the send uses your version);
  • Discard it so nothing is sent.

Notifications

When a draft is waiting, the platform notifies the responsible users — usually the enrollment assignee and the conversation assignee, falling back to administrators when no assignee is set. There are also reply received and enrollment failed notifications.

Safe degradation

If the Maestro is unavailable when the step runs, both AI assisted and AI full send the step's static fallback text. The cadence never stops because of an AI outage.

WhatsApp window respect

AI composition still respects WhatsApp's 24h window: outside it (WhatsApp Cloud), only an approved template (HSM) can be sent. Keep a template configured on the step.

Use cases

  • AI assisted for sensitive messages (billing, negotiation), where a human checks the tone before sending.
  • AI full for simple, low-risk touches, scaling without manual review.
  • Static with variables when the text is standardized and doesn't need AI.

Tips, limits & best practices

  • Start with AI assisted: review a few drafts to calibrate the intent before moving to AI full.
  • Always fill in the static fallback text — it's your safety net.
  • Write specific intents (goal, tone, language) for better drafts.
  • Make sure the responsible users receive notifications so drafts don't sit idle.

Troubleshooting

  • The draft stayed pending and no one saw it: check who the enrollment/conversation assignee is and the account's notifications.
  • The AI sent the static text: the Maestro was likely unavailable — degradation is the expected behavior.
  • The AI generated nothing: check that the Maestro is enabled and that the step has a defined intent.

See also