Hand off conversations to human agents

Wassla AI agents resolve routine questions on their own and escalate to your team the moment a conversation needs a person. You configure escalation rules in the agent's Personality tab, teammates take over from the shared Inbox with one click on the AI Active / Human Active badge, and internal notes carry context across the handoff without ever being shown to the customer.

How handoff works in Wassla

Every conversation in Wassla carries a handler_mode field that is either ai (the AI agent is replying) or human (a teammate has taken over and the AI is paused for that thread). The AI also has a built-in create_ticket tool it uses to log escalations into the Tickets queue when a customer reports a bug, asks for a refund, requests sales contact, or asks for a manager. Three controls work together to manage handoff:

  1. The agent's system prompt and escalation policy decide when the AI raises its hand
  2. The Inbox queue filters and the AI / Human toggle decide who is on the keyboard right now
  3. Internal notes and ticket comments carry context between the AI's reasoning and your teammates

1. Configure when the AI should hand off

The decision logic lives inside the agent's Personality tab. Wassla does not require you to write code or wire up a separate rules engine — you describe the escalation policy in plain language and the AI follows it.

  1. Open the Agents page from the sidebar and click the agent you want to edit.
  2. Open the Personality tab.
  3. In the system prompt, add an Escalation Policy section. Be specific. For example:
    • "Hand off to a human if the customer asks to speak to a manager, mentions a refund over 500 SAR, reports a payment failure, or uses angry language."
    • "After three turns without resolving the customer's question, open a ticket with priority normal and tell the customer a teammate will follow up."
    • "If the customer asks about enterprise pricing or asks for a demo, open a ticket of type sales and let them know the sales team will reach out within one business day."
  4. Save the agent. The change applies to new turns immediately — no redeploy is needed.

The AI calls its create_ticket tool whenever the policy fires. The ticket lands in the Tickets page with the conversation linked, and it also appears in the conversation thread as a system note so your team sees the escalation in context. The AI is instructed to open at most one ticket per distinct issue, so a follow-up message about the same problem updates the existing ticket rather than creating duplicates.

For sentiment-driven handoff, mention sentiment explicitly in the prompt ("if the customer sounds frustrated or angry, open a ticket and stop trying to resolve it yourself"). Wassla's background sentiment analyzer feeds into the AI's context, but the escalation decision is the AI's — driven by your prompt.

2. Pick up the conversation in the Inbox

The Inbox is the unified workspace where your team handles live customer conversations across WhatsApp, Instagram, Facebook, SMS, voice, and the web widget.

  1. Open the Inbox from the sidebar. The left column lists every active thread.
  2. Use the queue chips at the top of the sidebar to focus your view: All, AI (still being handled by the agent), Human (already taken over by a teammate), or Playground (test conversations from the agent playground, hidden by default). The Status toggle above the queue chips switches between Open and Closed.
  3. Click a thread to open it. The conversation header shows an AI Active or Human Active badge on the right.
  4. Click the badge to toggle handler mode. Switching to Human Active mutes the AI for that conversation — new customer messages sit in the queue waiting for your reply instead of triggering an automated response. Switching back to AI Active hands control back, and the agent resumes with the full transcript as context.
  5. Type your reply in the composer and press Enter to send. Your message goes out over the original channel (WhatsApp, Instagram, SMS, web widget) under your teammate name.

A useful shortcut: if the AI Active badge is still on and you type a one-off reply without explicitly taking over, Wassla sends that message and immediately releases control back to the AI for the next customer turn. This is handy when you only need to nudge a single message. Use the explicit toggle when you intend to own the conversation end-to-end.

For WhatsApp specifically, Meta only allows free-form replies within 24 hours of the customer's last inbound message. Wassla shows a window notice above the composer and disables the send button once the window closes — you can still reply with an approved template, and internal notes always work.

3. Use internal notes to brief the team

Internal notes are messages that stay inside Wassla. The customer never sees them. They are the standard way to brief a teammate before or during a handoff, ask a colleague for input, or leave a reminder for the next shift.

  1. In the composer, start your message with a single backslash (\).
  2. Wassla shows an amber "Note — only visible to your team" indicator above the composer so you can confirm the message will not be sent to the customer.
  3. Press Enter. The note appears in the thread with a distinct centered note style and remains visible to anyone on your team who opens the conversation later.

Notes work alongside the Tickets system. When the AI opens a ticket from a conversation, the ticket carries a link back to the source thread plus any notes the AI added during its reasoning. Your teammate can read the ticket, jump into the conversation, and reply with full context — no need to scroll through dozens of turns to figure out what happened.

4. Hand the conversation back to the AI

When you have resolved the customer's question or completed the escalation, click the Human Active badge to switch it back to AI Active. The agent picks up at the next customer turn and uses the full transcript — including your replies and internal notes — as context. There is no separate "close" step required to release control; closing the conversation as resolved is a separate action under the conversation header menu.

Tips

  • Keep escalation rules short and concrete. "Hand off on refund requests over 500 SAR" beats "be helpful and escalate when appropriate."
  • Use queue filters to split workload. Support specialists can live in the Human queue while a supervisor monitors AI for misfires.
  • Review the Tickets page weekly. Tickets the AI opened automatically are a free signal of where your knowledge base needs more coverage — see Train agents with a knowledge base to close those gaps.
  • Always set the agent persona to a real-sounding name. Customers respond better to "Layla from support" than to a generic "AI assistant", and the handoff feels less jarring when the teammate name appears in the same thread.