Ticket deflection

Ticket deflection has been the standard self-service metric for years. If a customer opens the help widget, reads an article or talks to a bot and does not then create a ticket, the enquiry is counted as deflected. Helpdesk platforms such as Zendesk, Freshdesk and Intercom report it out of the box, which is why it is the number most support teams already have.

The problem is what it does not measure. A customer who gave up, emailed a different address, phoned instead, or churned quietly is "deflected". A bot that makes reaching a human difficult scores well. Deflection therefore tends to overstate the value of a support AI and hides its failures. The better measure is auto-resolution rate, which only counts conversations where the problem was fixed and stayed fixed.

Deflection still has a legitimate use: as a capacity-planning input (how many contacts will the team not see) and as a leading indicator when paired with resolution and CSAT. Reported alone, it is misleading. Eazyware contracts on resolution and reports deflection as context.

• Support manager: it is the number your helpdesk already shows and your team is likely being judged on; understanding its blind spots matters before an AI project inflates it.
• CFO: headcount savings modelled on deflection will not materialise if deflected customers return through a costlier channel.
• Founder / CEO: vendor claims of high deflection are cheap to achieve and say little about customer experience.
• Product manager: a rising deflection rate with flat or falling CSAT is a warning sign, not a success.

Deflection exists because support teams needed a way to justify help centres and bots in terms of avoided tickets, and it is easy to compute from helpdesk data. It solves the reporting problem, not the customer problem. The trade-off is that optimising for it rewards friction: hiding the contact button, forcing article reads, closing chats early. Eazyware's position is that deflection is a capacity metric and should never be the success criterion for an AI support agent; we pair it with resolution, reopen rate and CSAT so that avoided tickets and solved problems are counted separately.

• SaaS help centre: measuring how many searches end without a ticket, then checking resolution via a follow-up prompt.
• E-commerce peak season: forecasting human queue size from expected deflection on order-status queries.
• Bank app support: tracking deflection on card and statement queries while monitoring call-centre volume for leakage.
• University student services: comparing pre- and post-AI ticket volume during admissions while checking reopen and walk-in rates.
• Logistics: measuring chatbot deflection on "where is my parcel" and reconciling with inbound call volume.

MetricA helpdesk metric you track and interpret, not a skill. Eazyware reports it as a secondary measure under Customer Service Agents, always alongside auto-resolution rate so that avoided tickets and solved problems are never confused.

Eazyware service that covers it: AI Customer Service Agents. Starting prices are on the pricing page.