Skip to content
Tutorials Center
My learning

Topics

Lesson 31 of 37 ยท Metrics and Frameworks

SOC Metrics: MTTD, MTTR and More

Measure a SOC properly: MTTD, MTTA, MTTR, dwell time, alert volume, false positive rate and SLA compliance, with medians, pitfalls and runnable calculations.

"How is the SOC doing?" is a hard question to answer without numbers, and an easy one to answer badly with the wrong numbers. Metrics let you show leadership the value of the team, find bottlenecks, plan staffing, and prove that changes (a new detection, a new playbook) actually help. They can also do harm: a team judged only on "tickets closed per hour" will close tickets quickly and carelessly.

This lesson defines the common measurements, shows how to compute them, and explains how to avoid being misled by them.

Time-based metrics

flowchart LR
    A["Attacker starts"] -->|"dwell time"| B["Alert created"]
    B -->|"MTTA"| C["Analyst picks up"]
    C --> D["Detected as incident"]
    D -->|"MTTC"| E["Contained"]
    E --> F["Recovered"]
    A -. "MTTD" .-> D
    D -. "MTTR" .-> F

The names vary by company. Always define start and end events in writing.

Metric Meaning Start event End event
MTTD mean time to detect How fast do we notice real incidents? Attacker's first action (or alert creation, if the start is unknown) Incident declared
MTTA mean time to acknowledge How fast does a human look at an alert? Alert created Analyst assigned or opens it
MTTC mean time to contain How fast do we stop the spread? Incident declared Containment done
MTTR mean time to respond/recover How fast to end the incident? Incident declared Closed or recovered
Dwell time How long was the attacker inside undetected? First attacker activity Detection

Be explicit: "R" in MTTR can mean respond, resolve, remediate or recover, and different vendors use different definitions. Pick yours and publish it.

Metrics that matter

Speed MTTA, MTTD, MTTC, MTTR
Quality False positive rate, escalation accuracy
Volume Alerts, incidents, backlog
Coverage Log sources, ATT&CK techniques

Speed alone rewards the wrong behavior.

1 / 2

Quality and volume metrics

  • False positive rate: share of alerts that turned out benign. High values mean noisy rules and tired analysts (alert tuning).
  • Alerts per analyst per shift and backlog size and age: are people overloaded?
  • Escalation accuracy: of the cases Tier 1 escalated, how many were real?
  • SLA compliance: percent of alerts handled within target time by severity.
  • Detection coverage: percentage of critical systems sending logs, and of relevant ATT&CK techniques with a working detection (ATT&CK coverage).
  • Automation rate: share of cases closed or enriched without hand work.
  • Reopen rate: incidents reopened later, a sign of hasty closure.

Mean versus median

Response times are skewed: most alerts take minutes, a few take days. One three-day outlier can drag the mean far above what most cases experience. Report the median and a high percentile (the 90th) beside the mean, or show a distribution.

Python Mean, median and 90th percentile

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The mean looks alarming, while the median shows that a typical alert is picked up in a few minutes. The outlier is real, though, and a good report explains it (no night shift) rather than hiding it.

Computing a small dashboard

Python Compute SOC metrics from case records

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Pitfalls

  • Goodhart's law: when a measure becomes a target it stops being a good measure. If closure time is rewarded, analysts close early. Pair speed with quality (reopen rate, escalation accuracy).
  • Vanity metrics: "we blocked 12 million events" says nothing about risk.
  • Missing denominators: 40 incidents is good or terrible depending on logging coverage and company size.
  • Averages hiding extremes: use medians and percentiles, and look at the slowest cases.
  • Comparing teams unfairly: severity mix, shifts and tooling differ.
  • Measuring what is easy: true detection quality (what you missed) cannot be counted directly; use exercises, purple-team tests and incident reviews.

Keep a short, stable set of metrics tied to goals: speed (MTTA, MTTC), quality (false positive rate, reopen rate), coverage, and one business-facing measure such as incident impact. Review the same numbers every month and annotate the changes you made.

Reporting

Different audiences need different views. Analysts want queues and aging. Managers want trends and SLA. Executives want risk: how long attackers stay undetected, which business areas are exposed, and what the money buys. Use charts for trends, a few headline numbers, and one sentence on what changed and what you will do next.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. Why report median and 90th percentile next to the mean?

    The median shows the typical case and the percentile shows the slow tail.
  2. 2. What does dwell time measure?

    Dwell time is the gap between the attacker's first action and detection.
  3. 3. A team is rewarded only for fastest ticket closure. What risk does this create?

    Goodhart's law: optimizing a single measure distorts behavior. Add quality measures such as reopen rate.
  4. 4. Which metric points to noisy detection rules?

    Many benign alerts mean wasted effort and alert fatigue; tune the rules.

Summary

  • Define each time metric's start and end events: MTTA, MTTD, MTTC, MTTR, dwell time.
  • Add quality, volume and coverage metrics, not just speed.
  • Use medians and percentiles, and always explain outliers.
  • Beware vanity metrics and Goodhart's law; report different views to different audiences.

Next: MITRE ATT&CK and Detection Coverage