Short answer
Deflection rate is the percentage of customer questions handled through self-service, such as a help center, chatbot or AI agent, without reaching a human agent. It is calculated as deflected contacts divided by total contacts, times 100.- Also called
- Ticket deflection, case deflection, support deflection, call deflection
- Counts as deflected
- Any contact that never reached a person
- Blind spot
- Customers who gave up count as successes
- Read it with
- Resolution rate, CSAT and repeat contacts
Deflection comes from the idea of turning a contact away from the support queue toward something cheaper: a help article, a chatbot, an AI agent. Deflection rate measures how often that happened. It is one of the most widely reported support metrics, and one of the easiest to misread.
How to calculate deflection rate
Divide deflected contacts by total contacts for the same period, then multiply by 100.
In October a Shopify store had 5,000 conversations start in its chat, and 3,100 of them never reached a person: 3,100 divided by 5,000, times 100, is a 62% deflection rate.
Three choices decide what the number means:
- The window. A customer who leaves the chat and emails an hour later was not deflected. Check for a follow-up contact within a few days.
- The starting point. Count conversations where the customer asked something, not every chat window that opened.
- Exclusions. Remove spam and test messages from both sides of the division.
Types of deflection
The same idea runs under several names.
- Ticket or case deflection. A customer reads a help article and does not submit the ticket. Measured as help center visitors who did not contact support.
- Chatbot or AI deflection. A conversation handled by a bot or AI agent without a person joining.
- Call deflection. Phone callers moved to a digital channel or self-service, such as a text link sent from the phone menu.
Measuring the quality of AI deflection
A deflection rate says how much an AI agent handled. It says nothing about how well. Four checks fill that gap:
- Repeat contacts. How many customers came back about the same issue within a few days.
- Satisfaction. CSAT on AI-handled conversations, compared with conversations people handled.
- Exits. How many conversations ended with the customer leaving mid-answer.
- Reading. A weekly sample of AI-handled conversations, read by someone who knows the policies.
In Aide, that reading has a record behind it: each reply shows the sources it was written from, and each customer message names the procedure Aide followed, so a reviewer sees why an answer was given, not only what it said.
Deflection is not resolution
A deflected contact is one that never reached a person. A resolved contact is one where the problem was solved. The two overlap, and the gap between them is where customers get lost.
| Deflection rate | Resolution rate | |
|---|---|---|
| Counts | Contacts kept away from people | Problems actually solved |
| A customer who gives up counts as | A success | A failure |
| Rewards | Fewer contacts reaching the team | Customers getting what they needed |
The practice this page rejects is optimizing deflection for its own sake. A store can raise its deflection rate by hiding the contact link, and every customer who gives up looks like a win. Resolution rate is the measure worth leading with, with deflection as a capacity number underneath it.
How to improve deflection without gaming it
Raise it by answering more questions, not by blocking more customers.
- Fix the help articles behind the questions customers ask most.
- Give the AI the store's live order data, so order questions get real answers.
- Make reaching a person easy. Customers who trust they can reach someone stay in self-service longer.
- Read the conversations that came back, by request type, every week.
In Aide, every night brings a list of verified knowledge gaps: questions the team answers that the help articles don't cover, each checked against the whole knowledge base and proposed with the text and the article it belongs in.
Where Aide fits
Aide's analytics count resolutions, not deflections: they separate the conversations Aide resolved on its own from the ones a person finished, show CSAT for Aide and for the team separately, and list handoff reasons by channel. Every number opens the conversations it counts. In Aide Chat, a shopper who clicks Talk to a person reaches the team at once. See Aide analytics.