Human Handoff and Escalation

What handoff means in this project

Human handoff in OrcaPulse is the point where the system stops letting AI or automated workflow behavior lead the interaction and instead gives control to an operator, agent, or transferred call destination.

In the current project, this already exists in two practical forms: social conversation takeover for inbox-style conversations, and AI call transfer for live calls that need a real person.

Conversation takeover and release

Social conversation handling already supports operator takeover. The social media controllers check whether a conversation has been taken over by a human agent and, if so, skip AI and auto-reply behavior for that lead.

Operators can also send manual replies, review full conversation history, mark conversations as read, and release a conversation back to AI later if needed.

• Takeover: stops AI and auto-reply logic for a social conversation that now needs human control.
• Manual reply: lets operators answer directly while preserving conversation history on the lead.
• Release: returns a conversation back to AI handling after the operator finishes intervening.
• Inbox-style visibility: teams can inspect full DM history before deciding whether to take over.

AI call transfer and escalation

AI calling also supports escalation. Assistants can carry transfer configuration, including a transfer number, transfer prompts, transfer message, and auto-transfer conditions. When enabled, the realtime call controller can detect transfer intent and move the caller to a human destination.

This means escalation is not only a manual operator workflow. It can also be part of the live call experience when the caller asks for a human or the configured transfer logic decides the conversation should be escalated.

• User-request transfer: phrases like asking for a human, rep, operator, or support can trigger transfer when enabled.
• Configured transfer destination: assistants validate that a transfer number exists before transfer can be used.
• Transfer event tracking: transferred calls are recorded in call events and reflected in end reasons and summaries.
• Fallback behavior: if transfer fails, the session can fall back instead of silently disappearing.

Operator controls and review surfaces

The project already gives operators places to monitor and intervene after a lead has entered a workflow. Review Leads and Lead Hub are the main operational surfaces, while timeline events and execution status make escalation decisions visible later.

There are also manual controls for stopping execution and cancelling recall flows, which matter when automation should pause and a human should take over instead.

• Review Leads: inspect execution progress and call outcomes.
• Lead Hub: review lead state and timeline history across workflows.
• Manual stop controls: stop one lead or bulk-stop execution when the workflow should not continue.
• Timeline history: confirm when assignment, pauses, failures, or transfer-style actions happened.

When to escalate in practice

The right moment to escalate depends on the channel and the workflow objective, but the current project already suggests a few strong patterns: move social DMs to a human when nuance or trust matters, transfer AI calls when the caller explicitly asks for a person, and stop automated follow-up when an operator should take over the relationship.

The best first implementation is usually narrow. Pick only the escalation moments that are obvious and easy for your team to review, then refine the triggers later.

• Escalate on explicit request: if the lead asks for a human, treat that as the clearest transfer signal.
• Escalate on high-value cases: let operators step in when lead quality or deal value justifies it.
• Escalate on failure patterns: stop automation if repeated retries, low-confidence outcomes, or confusing conversations appear.
• Keep the handoff visible: make sure your team can see the timeline and conversation state after the switch.

Next steps

After human handoff logic is clear, the next layer is deciding what should happen if no handoff is needed and automation should continue on its own for hours or days.

That usually means moving into follow-up automation and longer-running workflow sequences that keep leads warm after the first conversation or escalation point.