MTTA, MTTR and other incident metrics
What MTTD, MTTA, MTTR and MTBF measure, where each clock starts and stops, and how to use them without gaming them.
Published 30 September 2026.
Incident metrics measure how quickly problems are noticed, taken on and fixed. Used carefully, they show where the process is slow. Used as targets for individuals, they get gamed.
The metrics
| Metric | Measures | Clock starts | Clock stops |
|---|---|---|---|
| MTTD, mean time to detect | How good the monitoring is | The problem begins | An alert fires |
| MTTA, mean time to acknowledge | How well paging and escalation work | An alert fires | Someone acknowledges |
| MTTR, mean time to resolve | How long users are affected | An alert fires, or the problem begins | Service is restored |
| MTBF, mean time between failures | How often things break | One failure ends | The next begins |
The trouble with MTTR
The R in MTTR has at least four meanings in common use: repair, recover, resolve and respond. They measure different things. "Time to respond" is close to MTTA. "Time to resolve" can include follow-up work long after users were fine.
Pick one definition, write it down, and say where the clock starts. Otherwise two teams can report MTTR numbers that cannot be compared.
What MTTA tells you
MTTA reflects on-call setup most directly. A high or rising MTTA usually signals one of these issues:
- Notifications fail to reach people: wrong channel, phone on silent, or app not installed.
- Escalation waits are too long, or no escalation exists.
- Too many alerts make people slow to believe the next one. See alert fatigue.
Look at the distribution, not only the mean. A mean of 4 minutes can hide a handful of 40-minute acknowledgements at night. Those late-night delays are the ones that matter.
Using the numbers well
- Use medians and percentiles. Incident durations skew; one long outage dominates the mean.
- Measure the process, not people. A slow acknowledgement reflects the rota and notifications. It is not a mark against the person on call.
- Look at trends over quarters. Month-to-month noise is large when incident counts are small.
- Pair speed with volume. A falling MTTA with rising page count may mean people are exhausted and acknowledging everything instantly.
Beyond time metrics
Also track:
- Pages per on-call shift, and how many were at night.
- Share of pages that were actionable.
- Postmortem action items closed on time.
These say more about whether on-call is sustainable than any mean time does.
Most on-call and incident management tools report these metrics from their own data.