Auto-resolution rate

Auto-resolution rate counts a conversation as resolved only when the AI agent completed the customer's actual request (answered the question correctly, changed the address, issued the refund, rebooked the delivery) and the customer did not come back on the same issue within a defined window, typically seven days, through any channel. The denominator is all conversations the AI was offered, not only the ones it chose to handle, which prevents the number being inflated by routing hard cases away.

It is the counterpart to ticket deflection, which counts conversations that never reached a human regardless of outcome. Deflection can rise while customers grow more frustrated; auto-resolution cannot. It is also distinct from containment in voice, which measures whether the call was transferred, and from first-contact resolution for human agents, though the definition is deliberately aligned so that the two can be compared.

Measuring it honestly requires a feedback loop: reopen tracking across channels, sampled human review of "resolved" conversations, and a customer confirmation step ("did that fix it?") where practical. Without that loop, the metric drifts towards vanity.

• Support manager: it is the one number that tells you how much real work the AI is taking off the team, and it exposes topics the AI handles badly.
• CFO: cost savings only exist for resolved conversations; deflected-but-unresolved ones come back as a more expensive second contact.
• Product manager: a low rate on a specific intent often points at a product or documentation problem rather than an AI problem.
• Founder / CEO: it is the metric a board can trust, because it is tied to customer outcome and reopen data rather than to the AI vendor's dashboard.

Early support bots were judged on how many chats they absorbed, and vendors optimised for that: bots that stalled customers, looped them through articles and closed the chat when they gave up. Auto-resolution rate exists to prevent that failure by tying the metric to outcome and reopen data. The trade-off is that it is harder to measure: you need reopen tracking across channels, a review sample and an agreed definition of "resolved" per intent. Eazyware's stance is that resolution, not deflection, is the number to contract against, and we set up the measurement in the first weeks of any support-agent build.

• B2B SaaS support: percentage of "how do I" and configuration questions resolved with a correct, cited answer and no reopen.
• E-commerce: order-status, cancellation and return-initiation conversations completed end to end including the action in the OMS.
• FinTech: card-block, statement and dispute-status requests resolved inside the chat with policy-gated actions.
• Logistics: delivery reschedules and address corrections executed in the dispatch system without a human touch.
• EdTech: fee, deadline and access queries from students resolved against the current academic calendar.

MetricA tracked outcome metric, not a skill. Eazyware reports it, alongside reopen rate and cost per resolved conversation, as the primary success measure for Customer Service Agents and defines it per intent during scope.

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