Escalate complex conversations to a human

A financial advisory firm's WhatsApp number gets flooded with the same handful of questions — loan eligibility criteria, document checklists, processing timelines — but the moment someone asks something specific to their situation, a scripted chatbot just repeats the wrong answer or loops back to the menu, and the customer gives up before a human ever sees the message.

AI agent handoff answers the routine questions using the firm's own knowledge base, then recognises the moment a question goes beyond what it should answer — or the customer simply asks for a person — and hands off to a live agent with the full conversation history and a summary attached, so nobody has to explain their situation twice.

For a business like a financial advisory firm, the highest-stakes decision is scope enforcement, and it has to happen server-side, not merely as an instruction inside the AI's prompt. A prompted-only boundary can be talked around or misjudged by the model in an edge case, and an incorrect answer on a regulated topic — financial eligibility, legal standing, medical guidance — creates real liability, not just an unhelpful experience. Scope limits enforced in the backend, rejecting or redirecting anything outside a defined boundary regardless of what the model wants to say, are what actually protect the business here.

Handoff reliability matters just as much as accuracy. The moment a customer asks for a human, that request must be honoured immediately — Wavy never lets an automated responder deflect a direct request for a person, because a customer who asks twice, or gets another bot reply after asking once, loses trust in the system, not just that one conversation. Customers must also always be able to tell they're talking to an AI; the system never presents replies as coming from a human agent.

Every message the AI sends is still a billed conversation under WhatsApp's pricing, which means a responder that loops, misunderstands, or drifts off scope before finally escalating isn't just a poor experience — it's real money spent on a conversation that should have been shorter. Keeping the knowledge base current and the scope tight matters here too: a well-scoped responder that escalates promptly costs less to run than one that keeps trying and failing before finally handing off.