Resolution rate

What is resolution rate?

Resolution rate is the percentage of customer inquiries a support system closes completely, with the customer's problem handled, no transfer to another tier, and no reopen inside the review window. It applies to AI agents, human teams, and blended queues on any channel where a conversation has a defined end.

The arithmetic is one division; the disagreement is always about the numerator. Two teams running identical volume can report 45% and 80% because one counts every conversation that ended and the other counts only the tickets that stayed closed for 72 hours.

How resolution rate works

Resolution rate comes out of a four-stage loop: intake, attempt, outcome tagging, and the reopen window. Each stage can move the number several points while nothing changes for the customer.

Intake sets the denominator, so the way a team defines ticket volume decides what the rate is a share of; excluding spam, duplicates, and abandoned sessions quietly lifts every figure downstream. The attempt is the answer or the action itself: a policy explanation, a refund, an address change. Outcome tagging records what happened, and it is the weakest link, because a ticket closed by timeout and a ticket closed by agreement look identical in most reporting tables. The reopen window is the audit step: hold the closed ticket for 24 to 72 hours and treat anything that returns as a failure.

Worked through: 12,000 inquiries in a month with 8,400 closed and no transfer gives 8,400 divided by 12,000, or 70 percent. If 600 of those reopen inside 72 hours, the settled figure is 7,800 divided by 12,000, or 65 percent.

Two neighbouring metrics fall out of the same loop. Escalation rate counts what the first tier handed on, and first contact resolution counts the subset closed in a single interaction.

What counts as resolved and what does not

• Answered the actual question: the reply addressed the specific inquiry the customer sent, judged against the ticket text itself, which requires sampling by a human reviewer.

• No escalation: the conversation ended at the tier that received it, with no handoff to a supervisor, a specialist queue, or a callback.

• Action completed where an action was required: refunds issued, orders cancelled, addresses changed; quoting the policy leaves a request for a refund open.

• No reopen inside the window: the ticket stayed closed for a set period, commonly 24 to 72 hours, after which a return is treated as a fresh inquiry.

• Ended but unhandled: an abandoned chat or a timeout counts toward chatbot containment rate, which asks only whether a human joined the conversation.

Resolution rate vs deflection rate vs containment rate vs first contact resolution

Four numbers get reported interchangeably in the same board deck, and they measure four different things. Deflection rate counts the contacts that never reached a human. Containment rate counts the conversations a bot held to the end. First contact resolution counts the issues closed in a single interaction. Resolution rate counts the inquiries whose underlying problem was handled and stayed handled, which makes it the only one of the four that a customer would recognise as a description of their own experience.

If you are staffing a queue, deflection and containment tell you what arrived at the door. If you are answering whether support worked, resolution rate is the number, and the distance between the two is mapped in this comparison of deflection rate and resolution rate.

Why resolution rate matters for customer experience

When nobody measures resolution, teams manage what moves easily: handle time, first response time, tickets closed per agent. All three improve when an agent closes a conversation quickly and the customer writes back tomorrow, so a queue can look healthy while one problem is paid for three times.

Customers experience the failure as repetition. Explaining a billing error twice to two different people costs more goodwill than one slow reply, and repeat contact is the strongest driver of the feeling that a company is not listening.

The tradeoff is real. An agent tuned to resolve aggressively will act when it should have asked, issuing a refund outside policy or changing an account on thin evidence. Resolution rate rises and error rate rises with it, which is why serious teams pair the metric with a quality-review sample and a policy-violation count.

How is resolution rate calculated?

The formula is resolved inquiries divided by total inquiries, expressed as a percentage. The arithmetic is trivial, and every argument about resolution rate is an argument about the two inputs.

Fix three things before you compute anything: what enters the denominator, what qualifies as resolved, and how long the reopen window runs. Freeze those three in writing, because changing any one of them changes the reported number more than a quarter of genuine improvement will.

No standards body publishes a target resolution rate, and self-reported figures move with the denominator, so the comparison worth making is your own number across consecutive months under a frozen definition.

What can be priced is the work that failed resolution creates. The U.S. Bureau of Labor Statistics puts median pay for customer service representatives at USD 20.59 an hour (USD 42,830 a year, 2024), which means an agent closing six to ten contacts an hour carries roughly USD 2.06 to USD 3.43 of direct wage cost per contact before benefits, tooling, and supervision. Every reopened ticket pays that twice.

How AI agents change resolution rate

Retrieval systems answer questions. Resolution usually requires an action inside a system of record: issuing the refund, pausing the subscription, reshipping the order, updating the address. The ceiling on an AI agent's resolution rate is therefore set by its write access and its permission scope, not by the fluency of its prose, which is why two deployments on the same model report very different numbers.

The second mechanism is confidence thresholding. An agent that hands off whenever certainty drops protects accuracy and caps resolution; an agent tuned to attempt more raises resolution and pushes more errors into the review sample. That dial is a policy decision made by the support leader, and it should be documented.

The consequence is commercial. Resolution rate stopped being an internal quality statistic and became the unit deployments are judged on, a shift traced in this account of moving from deflection to autonomous resolution.

How to improve resolution rate

Coverage comes first: list the contact reasons by volume and mark which ones have a complete action path, including the authentication step. Reasons with an article but no action path are the visible gap.

Integration surface decides the ceiling. Read access to an order system answers questions; write access closes them, and each additional connected system converts a category of inquiry from explained to resolved.

Governance means one named owner of the definition, and an event log showing what the agent did on each ticket. SOC 2 Type II is the framework that bites here, because it forces evidence of change management and access control over the write permissions an agent uses to resolve; GDPR forces a lawful basis and data minimisation on the ticket text you sample to judge whether a resolution was real.

The operational constraint most teams miss is latency in the metric itself. Under a 72-hour reopen window, today's resolution rate is provisional for three days, so weekly reviews must compare settled periods only.

Resolution rate and the self-service stack

Resolution rate sits downstream of every self-service investment. Ticket deflection work, meaning help centre coverage and chat entry points, decides which inquiries reach an agent at all, and it changes the mix the rate is computed over: easy questions leave the queue, the residue is harder, and the rate can fall while support genuinely improves.

It also interacts with average resolution time. Keeping difficult cases and closing them yourself holds them in the queue longer, so a team raising resolution rate honestly should expect mean time to closure to rise for a period.

What does resolution rate mean in plain terms?

Think of resolution rate as the share of repairs where the car left the shop working and did not come back the following week. Counting the cars that drove away is easy. Counting the cars that stayed fixed is the number that tells you whether the shop is any good.

Without that second count, a support team can look excellent for a quarter and lose customers the entire time, because every closed ticket is recorded once and every return visit is recorded as fresh demand from a new person.

The tradeoff people underestimate is emotional as much as analytical. The honest version of this number is always lower than the flattering one, and it stays lower even as the team improves. Publishing it means giving up a figure that used to look good, in exchange for a figure you can actually act on.

Common resolution rate mistakes

Three patterns account for most inflated resolution rates.

Counting the end of a conversation as an outcome is the first. A session ends for four common reasons: the answer landed, the customer gave up, the browser closed, the timeout fired. Recording all four as resolutions moves the metric while nothing moves in the business.

Trimming the denominator is the second. Excluding the categories automation cannot handle, along with duplicates and anything routed straight to a specialist, shrinks the total the resolved count is divided by and lifts the percentage mechanically, with no change in customer outcomes.

Reporting before the reopen window closes is the third. A dashboard showing this morning's resolution rate is showing a figure that cannot be final for another day or three, and comparing a same-day number against last month's settled number compares two different measurements.