Human Handoff for Website AI Chatbots: A Spec You Can Actually Build From

When creating a chatbot for a business website, it's crucial to design the handoff process correctly. This is where leads are won or lost, as the visitor who asks something the bot can't answer is usually the one with a real, specific need. The handoff spec starts by deciding what triggers a handoff before any coding begins.

Five trigger reasons are identified and defined as named reasons (USER_REQUEST, LOW_CONFIDENCE, RESTRICTED_TOPIC, BUYING_SIGNAL, LOOP), which will be used for routing and reporting. Next, minimal context is collected before the handoff to avoid visitor frustration. Only the contact method, a one-line summary of the problem, and consent information are requested.

The business hours are taken into account when crafting the handoff message, ensuring the visitor knows the expected response time. A standardized handoff payload is created, including the conversation ID, reason, summary, visitor information, page URL, UTM data, and retrieval score.

The handoff payload is then routed based on the identified reason, with different actions defined for each reason. For instance, BUYING_SIGNAL conversations are routed to a CRM deal alert, while RESTRICTED_TOPIC conversations are sent to a named owner. The bot informs the visitor that a person has taken over the conversation, when they can expect a reply, and where the reply will arrive.

To measure the effectiveness of the handoff process, weekly tracking is done, focusing on handoff rates, response times, and the top LOW_CONFIDENCE questions, which help identify areas for content improvement. This data-driven approach ensures the chatbot and the human support team are working together efficiently to meet the visitor's needs.