The share of questions answered by self-service — help center, automations — before they become an agent ticket.
Deflection is an industry concept, not a single feature. It measures how often a customer gets a complete answer without an agent reading or replying. The mechanism varies — a searchable help center on your own domain, an in-chat article suggestion, an automation that answers order-status questions, or an opt-in AI reply — but the test is the same: was the issue resolved before it landed in the agent queue?
Why it matters
Every deflected question is one an agent never has to type. Because Cherryrise prices per agent, deflection lets a small team absorb growing volume without growing headcount. It also speeds up the customer: a good help-center article or a canned response resolves a question in seconds rather than waiting for a reply. The goal is not to hide humans — it is to reserve human time for the questions that actually need judgment.
How deflection actually works
- Search-first content. Customers find answers themselves through a help center indexed by search engines and surfaced in the chat widget.
- Suggested articles. Before a chat or form submits, relevant articles appear; some are read and the conversation never starts.
- Automated answers. Rules and automations respond to predictable, structured questions — order status, password resets, business hours.
- AI assist. An opt-in, admin-gated model can draft or send answers grounded in your own content. See AI assist.
How to measure it
A common formula is deflected sessions divided by total help-seeking sessions — for example, help-center visitors who left satisfied versus those who then opened a ticket. Treat the number cautiously: a customer who reads an article and still files a ticket was not deflected, and a customer who gives up in frustration is a false win. Pair the rate with downstream signals like CSAT and reopen rate so you are not optimizing deflection at the cost of resolution.
Common pitfalls
The biggest mistake is burying the human path. If an e-commerce store wraps every chat in a chatbot the customer cannot escape, deflection looks high while frustration climbs and CSAT falls. Keep "talk to a person" one click away on every self-service surface. The second mistake is stale content: deflection depends on articles being accurate, so treat the help center as a living asset, not a launch-once project. For a deeper walkthrough, see how to reduce ticket volume without hiding support.