How is Waaru's handoff different from a shared inbox on Wati or AiSensy?

Waaru loads the full transcript, an AI summary, the contact record, and internal notes before the agent reads the first message. Most BSPs drop the customer into a shared inbox with no summary and no contact context, so the agent restarts from scratch.

When does the AI hand off?

Three triggers: the AI's confidence drops below a configurable threshold, the customer explicitly asks for a human, or your routing policy escalates the intent (complaints, high-value transactions, named VIPs).

Can I route handoffs to a specific team or agent?

Yes. Routing rules support intent, tag, VIP segment, language, business hours, and team. The conversation lands with the right human, not just any human.

Does the human's resolution improve the AI?

On the Scale plan, yes — every resolution is indexed and used as context for the next time the same intent appears. The AI handles the same question alone the next time, without manual retraining. Founding members get this from day one.

Can multiple agents see the same conversation?

Yes — the inbox is a team inbox. Agents see assignments, status, and live typing indicators. Internal notes are visible to the team but not the customer.

Is conversation data encrypted?

Yes — every message is stored encrypted at rest. Per-workspace isolation. No cross-tenant access. DPDP Act 2023 compliant.

What does HITL cost on Waaru?

Team inbox and handoff are included from the Starter plan. Self-learning HITL (resolutions teach the AI) is included from the Scale plan — and free during founding-member access on every plan.

WhatsApp human in the loop

Human-in-the-loop handoff with the full picture, not a cold transfer.

The wrong way to do human handoff: the AI gives up, drops the conversation into a shared inbox, and a human starts from scratch with no context. The right way: the AI escalates with a one-paragraph summary, the agent sees the full transcript and the contact's record, and the human's resolution teaches the AI for next time.

In one line

WhatsApp human-in-the-loop (HITL) is the handoff system between an AI agent and a human agent. Done right, the human gets full conversation history, an AI summary, and the contact record on handoff — and the resolution teaches the AI to handle the same intent alone next time.

No credit card. No commitment.

Full transcriptLoaded on handoff

AI summaryGenerated for every escalation

Self-learningResolutions index automatically (Scale)

0%Markup on Meta fees

The problem

Why most 'handoff' is a cold transfer.

On most BSPs, the AI's confidence drops, the conversation lands in a shared inbox, and the customer has to repeat themselves. The agent doesn't know what the AI tried, what the customer wanted, or what they were promised. The resolution rarely makes it back to the AI — so the same question gets escalated next week, and the week after.

No summary — the agent reads the whole transcript before replying, every time.
No contact record — past conversations, preferences, and orders aren't loaded.
No feedback loop — the human's correct answer doesn't teach the AI to handle the same intent alone next time.
Cold inbox — customers wait while staff figure out what the AI was trying to do.

Chapter 01

When the AI escalates, the human sees the full picture.

Every Waaru handoff loads four things into the team inbox before the agent reads the first message: the full conversation transcript labelled by speaker, an AI-generated summary of the situation in one paragraph, the contact's record (past conversations, tags, last order, preferences), and any internal notes from prior sessions. The agent picks up at exactly the right point. The customer never repeats themselves.

Chapter 02

AI summary in one paragraph — the situation, not the transcript.

The AI summary is not 'here are all the messages'. It is a one-paragraph synthesis: what the customer wants, what's been tried, what's blocking, and the AI's best guess at the right next action. Most agents read the summary, reply, and never need to scroll through the transcript. This is the difference between a one-minute handoff and a five-minute handoff. At ten escalations a day per agent, that compounds.

Chapter 03

Routing rules — the right human, not just any human.

Set routing rules per intent, per tag, per VIP segment, per language, per business hour, per team. A complaint goes to the senior agent; a Hindi conversation goes to a Hindi-speaking agent; an enterprise contact goes to their named account owner. The agent who picks up the conversation is the right one — not the next available.

Chapter 04

Status, notes, and audit trail.

Conversations carry a status (open, pending, resolved) and a thread of internal notes the customer never sees. Notes are timestamped, attributed, and searchable. Every reply, status change, and assignment is logged — if a customer complaint reaches you in two months, you can reconstruct exactly what happened.

Chapter 05

Self-learning — the human's resolution teaches the AI (Scale plan).

On the Scale plan, every human-resolved escalation is indexed automatically. When the same intent appears in a future conversation, the AI handles it alone using the resolution as context. Over the first six months, customers typically see the AI deflection rate climb by 1–2 percentage points per month without any manual retraining.

FAQ

Questions worth answering.

Human-in-the-loop is the handoff system between an AI agent and a human agent on WhatsApp. The AI handles routine messages, escalates when it hits a limit, and the human picks up with full context. Done right, the human's resolution teaches the AI for next time.

Keep reading.

Ready to ship a WhatsApp flow that doesn't break?

Apply for founding access. No credit card. Cancel anytime.

See pricing