MTTR meaning: recovery, response, repair and resolution

Understand four common MTTR definitions and MTTA, calculate each interval, and read incident reports without confusing acknowledgement, restoration and corrective work.

Best practices
Abstract illustration of incident signals connected to handling stages

MTTR can mean mean time to recovery, respond, repair or resolve. Before using a number such as “MTTR: 10 minutes,” ask which two timestamps it measures. An average from incident creation to ticket closure answers a different question from an average covering the actual service outage.

These terms vary between products and organizations. The definitions below give this article a consistent set of intervals; they are not a universal incident-data schema. Atlassian’s definitions and PagerDuty’s definitions already differ in what they call resolution time.

Recovery: how long the service was affected

Here, mean time to recovery runs from the failure to the point when normal service is restored. For the selected incidents:

Mean recovery time = sum of recovery durations / number of incidents

Suppose two outages in a week have a combined duration of 20 minutes. Their mean recovery time is 10 minutes. That is an illustrative calculation, not a benchmark or a target every service should meet.

Recovery time covers detection, notification, acknowledgement, diagnosis and restoration. A high value tells you that the overall path is slow; it does not identify the slow stage. Delayed notification and a difficult repair can produce the same total.

Response: check where the clock stops

This article follows Atlassian’s use of mean time to respond: the average from the first alert received to service restoration. To compare a report labelled “response time” with this measure, check its start and end conditions.

Under the definition used here, sum the alert-receipt-to-restoration intervals and divide by their incident count. Subtracting that mean from mean recovery time gives the average failure-to-alert-receipt interval only when both calculations use the same incidents, complete timestamps and the same restoration endpoint. This interval can include both detection and notification delays.

Consider one hypothetical timeline: impact starts at 10:00; an alert triggers at 10:02; its notification arrives at 10:03; someone acknowledges at 10:04; repair starts at 10:05; service returns at 10:12. Recovery takes 12 minutes, response-to-restoration takes 9, repair takes 7, and trigger-to-acknowledgement takes 2. These are four intervals within one incident.

Better alert routing and backup on-call coverage can shorten the wait before work starts. Runbooks can help responders diagnose and restore service. A runbook documents the response steps, roles and follow-up work; it does not remove the need to record when those steps happened. Acknowledgement ends one stage, while coordination and waiting for repair may continue.

Repair: time spent on the repair interval

Here, mean time to repair starts when repair work begins and ends when the fix is complete and restoration has been verified. Add those durations and divide by the corresponding incident count. Waiting before work starts is outside this definition.

Decide whether diagnosis and verification are included. Software diagnosis and repair often overlap, so including both can be practical if the rule is explicit. Other reports may start at fault detection or record these stages separately. Elapsed repair time also differs from the sum of every responder’s working hours. Use it to find obstacles in diagnosis, implementation or verification, rather than to rank individual engineers.

Resolution: define the agreed completion condition

For this article, mean time to resolve runs from failure to completion of the agreed corrective work. Restoration might come from a restart, while a permanent code change arrives later. Finishing that change reduces a known risk; it does not guarantee that a similar incident can never recur.

Some platforms instead stop at the incident’s resolved status. Specify whether follow-up work is included and whether the clock uses elapsed time or business hours. With a consistent incident cohort and clock, the difference between resolution and recovery means shows the average interval after restoration. Investigate review, development, testing and release schedules before interpreting a long interval as unfinished work.

MTTA: acknowledgement is a separate event

In this article, mean time to acknowledge (MTTA) runs from alert trigger to the on-call responder’s recorded acknowledgement. Sum those intervals and divide by the corresponding incident count. Other systems may start at incident creation; PagerDuty, for example, documents that basis and requires an explicit acknowledgement in its team reporting.

A rising MTTA may point to alert fatigue, failed delivery, incorrect ownership, gaps in coverage or missing acknowledgement records. Check notification history and the on-call schedule as well as alert volume and false positives. Acknowledging an alert does not prove that repair has started.

Match the interval to the question

Metric used here Start → end What it helps explain
Recovery Failure → service restored Overall restoration time
Response First alert received → service restored The path after the alert reaches the team
Repair Repair starts → fix and restoration verified Diagnosis, repair and verification within the agreed scope
Resolution Failure → agreed corrective work completed Follow-up beyond immediate restoration
MTTA Alert triggers → acknowledgement recorded Routing, delivery and on-call acknowledgement

Build the report from recorded events

Guance Incident Center records incident triggers, status changes, assignee changes and escalation notifications in its operation timeline. Incident details also provide current-cycle and cumulative MTTA/MTTR statistics. Those records help explain what happened during handling; first map the platform’s metric definitions to the report your team needs. Guance incident details

A status timeline cannot automatically supply the moment users first experienced impact, the start of repair work or the deployment of a corrective change. Fill missing milestones from service monitoring, deployment records or the postmortem. Logs and APM can help establish impact and restoration evidence, but cannot reconstruct a timestamp that was never recorded.

Questions that come up in reviews

Do we need all five metrics? Start with recovery time and acknowledgement if those address the team’s immediate problem. Split the remaining stages when the distinctions would change a decision.

What is a good MTTR? Use the service’s tolerance for disruption and its SLA commitments. Compare the same severity, service and timing rules, and show the sample count. A lower monthly average can reflect a different mix of incidents. Inspect individual durations and useful percentiles; a P95 from very few incidents is unstable. Report ongoing incidents separately so a mean over completed incidents does not hide them.

Does a high MTTA mean the on-call engineer was slow? The number alone cannot establish that. Check delivery, ownership, coverage, alert load and missing records first.

Can recovery and resolution be combined? A small team can use one completion milestone, provided it recognizes what disappears from the report: the gap between restoring service and completing follow-up work. Keep corrective actions separately trackable when they remain open after restoration.

Sources

Get a tailored plan

Contact us

Join the community

Scan with WeChat
to join the community

Try Guance

Start online and pay only for what you use.

Get started

Choose a Guance plan

Code hosting