Skip to content

Guide, read 10 Nov 2026

Measure an MDR service by what it did, how fast and what it missed, using definitions you fixed in advance

Why do published MDR figures not transfer?

Providers publish speed figures, but each defines its figure its own way. Expel publishes 5 minutes mean time to detect, 13 minutes to respond on high severity and 14 minutes to remediate. SentinelOne says Wayfinder MDR mean time to detect averages 3.3 minutes across customer environments. Sophos cites an average of 89 seconds from alert to automated response where AI can act alone. CrowdStrike states a median time to contain of 1 minute, with a footnote defining it from detection to containment on an endpoint. ReliaQuest says GreyMatter contains threats in under 5 minutes.

These are statements by the providers, read on 10 Nov 2026, and listed alphabetically. They differ in statistic (mean, median, average), in the start point and in the case mix. Do not set one beside another. Measure your own cases instead.

Five measures that work

Choose measures you can compute from the provider's case export and your own records.

  • Coverage: percent of assets and log sources reporting, checked weekly against your inventory.
  • Time to first action: from the alert time in your tool to the first analyst note or action on your case, by severity.
  • Closed-alert quality: sample ten alerts closed as benign each month and check them yourself. NIST's glossary defines a false positive as incorrectly classifying benign activity as malicious; the sample tells you whether closures were right.
  • Escalation quality: for each escalation, was the context enough to act, was the recommendation correct, and was anything repeated?
  • Change handling: days from your request to a detection change, exclusion or new source being live.

Resolved without you is not the same as resolved well

Sygnia states that it resolves 95% of alerts without involving the client's team. A figure like that tells you the share handled by the provider, not whether the handling was right. Pair any such number with the sampling check above. A low escalation count may mean a quiet month or a blind spot; the sample tells you which.

Track the reverse as well: alerts you found yourself that the provider did not. Keep a log of every finding from your own tools, scanners and staff, and ask the provider each time why the service did not raise it.

Mapping to ATT&CK

MITRE's ATT&CK is a knowledge base of adversary tactics and techniques. Ask the provider to show which techniques its detections cover for your sources, and compare that with the techniques most relevant to your sector. Treat the map as a coverage question and not as a quality grade: a technique with a rule is not a technique that is always caught.

MITRE's ATT&CK Evaluations has a Managed Services domain whose published rounds include OilRig (published 22 Nov 2022) and menuPass with ALPHV BlackCat (published 18 Jun 2024). If a provider cites a round, read the round's own publication for what was tested and for which service configuration.

The monthly review

Hold a monthly 45-minute review with the same agenda: the five measures, three sampled closures, every action taken, every exception to the authority matrix, open change requests and a decision list. Keep the minutes. At 6 and 12 months compare the trend with the contract commitments, including any credit you are owed. Record changes in your environment as well, since a new cloud account or a merger changes what you should expect.

Sources

  1. NIST CSRC glossary, false positive, read 10 Nov 2026
  2. MITRE ATT&CK, FAQ, read 10 Nov 2026
  3. MITRE ATT&CK Evaluations, adversaries list (Managed Services domain), read 10 Nov 2026
  4. Expel, SOC page, read 10 Nov 2026
  5. SentinelOne, MDR page, read 10 Nov 2026
  6. Sophos, MDR page, read 10 Nov 2026
  7. CrowdStrike, Falcon Complete page, read 10 Nov 2026
  8. ReliaQuest, home page, read 10 Nov 2026
  9. Sygnia, MDR page, read 10 Nov 2026

Last reviewed 10 Nov 2026.