Chatbot Deflection

Chatbot Deflection is a primary goal of Inbound Support AI. It reduces the load on your human team and provides customers with instant solutions to their problems. Separate contained contacts from unresolved escalations in reports. See Ticket deflection chatbot.

Chatbot deflection is a customer request resolved by the AI agent without a person touching it. The deflection rate is the share of conversations that end that way. It is the metric most support teams use to size the savings from automation, and the one most often measured badly.

Counting it honestly. A conversation counts as deflected only when the customer confirms the answer, or does not reopen the issue (by any channel) within a set window, usually 24 hours. A customer who gives up and emails the next day is not a deflection.

What deflects well. Order status, hours and location, policy questions, password and account basics, appointment changes and FAQ lookups: high volume, low risk, answerable from documents or a system lookup.

What should not be deflected. Billing disputes, suspected fraud, complaints with strong emotion, regulated advice and anything the agent cannot ground in approved content. The right outcome there is a fast, well-documented handoff.

Deflection vs containment. Containment counts conversations that never reached a person regardless of outcome; it rewards agents that stonewall. Deflection as defined above rewards agents that resolve.

Example. An ecommerce brand measures tickets per 1,000 orders before launch, deploys an agent on web chat and WhatsApp connected to Shopify, and tracks resolved-without-reopen conversations by intent for 30 days. Order-status and returns-policy intents deflect; damaged-goods claims route to a person with the order attached. Botcadence reports deflection by intent so the number is checkable; see how AI chatbots reduce support costs.