Cyber resilience metrics measure whether an organization keeps its critical services running through a disruption and restores them in the right order. Most programs time the technical clock, from first alert to restored system, while leaving the human decisions in between unmeasured. Those decisions can also be measured. This guide explains how.
What Are Cyber Resilience Metrics?
Cyber resilience metrics are measurements, or values computed from measurements, that show how far a disruption pushed a critical service below its minimum acceptable level, how long it stayed there, and how well the people responsible decided their way back.
Reflex Security publishes this article and sells crisis simulations that produce some of the data these metrics require. The definitions below also work with a spreadsheet, a scribe, and a wall clock.
The vocabulary comes from engineering. MITRE wrote its 2018 report, Cyber Resiliency Metrics, Measures of Effectiveness, and Scoring, for systems engineers and program managers. That origin explains why so many resilience metrics for cyber systems describe a performance curve: a disruption drops performance, recovery restores it, and the metric reads time against level. The report also separates three terms that dashboards tend to blur.
Table 1. Metric, measure of effectiveness, and score
Term | What it is (Bodeau et al., 2018) | Food processing example |
Metric | "The result of a process or method for measuring, evaluating, or comparing similar objects" | Minutes the pasteurization line ran below its minimum essential output |
Measure of effectiveness (MOE) | A metric used to judge whether a specific change achieves its resilience goal. MITRE adopts a definition in which an MOE is "an indicator used to measure a current system state, with change indicated by comparing multiple observations over time" | Whether a new escalation path shortened escalation latency across three exercises |
Score | A summarized judgment. MITRE's Situated Scoring Methodology for Cyber Resiliency gives "a simple relative measure" that depends on stated priorities and threat assumptions | One resilience score for the plant network, valid only under its stated threat model |
The order matters for everything below. A metric is a number you can collect today. It becomes an MOE only when you compare it before and after a change, so a single exercise cannot produce one. A score compresses many judgments into one figure and is only as sound as its assumptions.
The Cyber Resilience Metrics Every CISO Should Track
The key metrics for measuring cyber resilience share one property: each can come back red in a quarter the organization handled badly. A metric that cannot fail is decoration.
NIST SP 800-55 Vol. 1, published in December 2024, gives every measure a standard record. Its fields include the goal, the measure, its type, a formula, a target, implementation evidence, how often data is collected, the responsible parties, and the data source (Schroeder et al., 2024). I keep three of those fields in the table below. Formula, data source, and owner decide whether two people computing the same metric get the same number.
The eight rows are my proposals, not industry standards. I group them by the four goals MITRE's framework names: anticipate, withstand, recover, and adapt. The owners assume a food and beverage processor whose plants run on operational technology. OT below means those plant control systems. Substitute your own titles.
Table 2. Proposed cyber resilience metrics. Author-proposed definitions, not benchmarks.
Metric | Formula | Data source | Owner | Goal |
Declaration latency | declaration time − first-signal time | Alerting platform and incident record | SOC lead | Anticipate |
Escalation latency | time the named decision-maker joins − declaration time | Paging tool and bridge log | Incident commander | Withstand |
Decision completeness | required decisions made ÷ decisions the playbook requires for the scenario | Decision log checked against the required-decision list | Crisis management lead | Withstand |
Cross-function handoff count and delay | count of handoffs between functions; median of (receipt time − send time) | Bridge transcript and chat export | Incident commander | Withstand |
Critical-service time-in-degraded-mode | sum of minutes each critical service ran below its minimum essential level | Plant historian, service monitoring, and operations log | Plant operations manager | Withstand |
Restore-order accuracy | services restored in the agreed priority order ÷ services restored | Recovery runbook compared with the restore log | IT/OT recovery lead | Recover |
Corrective-action closure rate | actions closed by due date ÷ actions due in the period | After-action report action register | Security program manager | Adapt |
Repeat-finding rate | findings in this run also raised in an earlier run ÷ findings in this run | Findings register across exercises | GRC or metrics lead | Adapt |
Six of the eight rows measure what people do. That ratio is deliberate. Detection and response timing appears once, as declaration latency, because the clock that matters here is the one that ends when a person calls it an incident.
In NIST's four measure types (implementation, effectiveness, efficiency, and impact), most rows are effectiveness and efficiency measures. The board section below turns two of them into impact. NIST also argues for restraint, at least at first: with quantitative metrics, “less is often more.” Begin with no more than eight. If the available data cannot support one of them, record the missing data as a finding. Cyber Resilience Assessment: How to Measure Your Readiness covers the assessment blind spots that commonly explain it.
Beyond MTTD and MTTR: Measuring Organizational Response
MTTD is mean time to detect, which NIST SP 800-55 Vol. 1 defines as “the average amount of time that a problem exists before it is found.” MTTR, in the same document, is mean time to recovery: “the average amount of time it takes to recover from a product or system failure.” Other teams use the same letters for respond, repair, or resolve.
These measures can describe a defined technical process when the start point, stop point, and event class remain consistent. They become unreliable readiness indicators when averaged across incidents with different causes, systems, severity, and business consequences. Ticket timestamps can introduce another problem when records are created manually or after the work begins.
For that reason, this article does not treat MTTD or MTTR as complete measures of organizational readiness. The business crisis depends on the decisions between detection and recovery: who engaged, what was decided, which handoffs occurred, and which required tasks remained incomplete. Any answer to how to measure cyber resilience needs to record those events directly.
Table 3. The technical clock and the organizational layer
What the clock captures | What it misses |
When the alert fired | When someone with authority called it an incident |
When the system came back | Whether services came back in the order the business needed |
How long containment took | Who decided to stop the line, and how long that decision waited |
Ticket timestamps | Handoffs made by phone and never logged |
The right-hand column is the organizational-response layer. The next section explains how to measure it, while Moving Beyond Tabletop Exercises explains the case for rehearsing it under pressure.
Measuring Escalation, Decision-Making & Cross-Functional Coordination
The standard objection is that human response is subjective. Part of it is. Most of it is not. NIST lists consistency among the characteristics of a measure, meaning results that "remain independent of the individuals or entities conducting them." A timestamp passes that test, and so does a missing decision. A judgment about decision quality does not, so it gets its own column and a named reviewer.
Take a hypothetical food and beverage processor in peak season. Ransomware reaches the plant network. Partway through, the pasteurizer's temperature readings in the plant historian stop matching the paper chart recorder, which puts a contamination risk on everything produced since. Quality, operations, legal, and communications now share one question: stop the line, hold the affected lots, or both. The plant's own cold-chain procedure gives chilled finished goods 4 hours outside monitored storage, a hypothetical figure, and the refrigeration controls sit on the same network. That is the clock that matters.
Five steps turn that scene into data.
Fix the event vocabulary before the exercise. Define in writing what counts as first signal, declaration, decision-maker engaged, decision made, and handoff received. This scenario has two first signals. The ransomware alert starts declaration latency, and the historian mismatch starts the quality clock on the lot-hold decision. Every formula in Table 2 depends on these definitions holding still.
Write the required-decision list. Name each decision the scenario must produce and its owner: line stop (plant operations manager), lot hold (quality lead), customer notice (communications), and records preservation (counsel). This list is the denominator for decision completeness.
Keep a decision log. One line per decision: time, decision, owner, evidence used, option rejected. For example: "14:06 · hold lots 4471 to 4479 · quality lead · historian gap from 13:12 · rejected: release on chart-recorder data alone." Meeting minutes summarize the discussion, while the decision log records the commitment made.
Map who spoke to whom. From the bridge transcript or chat export, draw every exchange between functions with its timestamp. Handoff count and delay come from this map. So does the disagreement: in the hypothetical, operations wants the line running to protect peak orders while legal wants a hold to limit recall exposure. The map shows whether quality heard both positions before deciding.
Run the missed-required-decision check. After the exercise, compare the log with the list. Any required decision with no entry is a miss, however thorough the discussion felt. A lot hold placed after the cold-chain window closed is late. Report late separately from missed.
Behavioral scoring sits one level up. Ron Dilley introduced the VECTOR model (Velocity, Evidence, Command, Teaming, Optionality, and Resilience) when he presented the results of the Hacker Halted simulation at the Global CISO Forum. It is one example of a behavioral model built on the same timestamped record. Used to rate a team, it would sit in Table 1's score row, above the metrics. Reflex is one source of that record. A disciplined scribe is another.
Which Cyber Resilience Metrics Belong in the Boardroom?
Directors do not need eight rows. They need the few cyber resilience metrics that translate into business exposure and are reported consistently each quarter. Four belong in the board view.
Time critical services ran degraded. Minutes below minimum essential output, per critical service, with real incidents and exercises in separate columns. What the board should ask: which service ran degraded longest, and was that inside the tolerance we approved?
Decision latency trend. Median escalation latency across the last four comparable runs. What the board should ask: is the right decision-maker engaging sooner under comparable conditions, and where is that not improving?
Open high-severity corrective actions. Count, age, and owner. What the board should ask: which are past due, and what exposure stays open while they wait?
Exercise coverage of critical services. Critical services exercised in the last 12 months ÷ all critical services. What the board should ask: which critical services has nobody rehearsed?
Measuring cyber resilience maturity as a trend, not a label, means showing these four lines moving over time. NIST SP 800-55 Vol. 1 makes the case against the label itself. When an assessment says an organization is at "level 3," the document notes that "3" "represents a set of qualities rather than a numerical value." A time series can instead show whether performance is improving.
Turning Crisis Simulations Into Measurable Readiness Data
A metric becomes useful as a series only when the formula and measurement conditions remain comparable across runs. MITRE makes the point directly: metric values “tracked and compared over time for the same organization, mission or business function, or system can be useful to identify trends.” Beyond that, “comparison becomes more challenging.” An organization’s own controlled baseline is more defensible than an external average.
Treat the first run as the baseline and resist grading it. Then decide what stays fixed. If the scenario repeats exactly, the team learns the answers and the series measures memory. If everything changes, nothing is comparable. Hold the required-decision list constant (line stop, lot hold, and customer notice) and vary the entry point, timing, and unavailable participants. Reflex generates five customized scenarios in 10 to 15 minutes, which makes that variation practical. AI-Powered Tabletop Exercises for Incident Response explains how the adaptive scenarios work.
Comparable trends also require regular testing. NIST notes that organization-level measures require “the processes on which the measures depend” to be “consistent and repeatable.” One annual tabletop provides one observation, while quarterly runs provide four observations within 12 months.
Table 4. Illustrative run-over-run series. Hypothetical values for the food processing scenario, same six required decisions each run. Not benchmarks.
Metric | Run 1 (baseline) | Run 2 | Run 3 | Run 4 |
Declaration latency (minutes) | 46 | 38 | 31 | 24 |
Escalation latency (minutes) | 41 | 40 | 22 | 19 |
Decision completeness | 3 of 6 | 4 of 6 | 4 of 6 | 6 of 6 |
Repeat-finding rate | n/a | 50% | 33% | 20% |
Read Run 3. Escalation latency fell by almost half after one change, a named deputy for the plant operations manager. That before-and-after comparison is what turns a metric into a measure of effectiveness.
Two adjacent jobs sit outside this section. Cyber Resilience Testing: How CISOs Can Prove Resilience covers how to determine whether test objectives were met. Cyber Crisis Simulation: The CISO’s Guide to the Next Generation of Cyber Exercises explains how simulations capture decisions. Turning the series into organizational change starts with measuring tabletop results.
From Simulation Evidence to a CISO-Ready View of Incident Readiness
Reflex is one data source for the series. An exercise commonly runs for about an hour, with duration configurable. Reflex records participant actions, decisions, communications, and Investigation Console activity. When the exercise ends, it produces an After Action Report (AAR) with timestamped findings and action items assigned to owners and deadlines.
That record feeds most of Table 2:
The timeline and transcripts supply escalation latency, decision timing, and handoff count and delay.
The decision record, checked against your required-decision list, supplies decision completeness.
AAR action items, carried from run to run, supply corrective-action closure and repeat-finding rates.
Time-in-degraded-mode and restore-order accuracy still come from your own service monitoring and restore logs. An exercise can rehearse the restore order. It cannot measure your plant's real output.
The CISO view is one page: the eight rows, the last four runs of each, a direction for each, and the open high-severity actions with their owners. Every number on it should trace back to a timestamp. In Reflex, the Digest step produces the after-action evidence that page draws on, with each finding traced to the moment in the simulation it came from.
Pick one scenario, write its required-decision list, and run it twice. The second run provides the first point of comparison.
See what your team does under real pressure.
Put your readiness to the test. Sixty minutes, your real environment, no slides.
Book A demo
Frequently Asked Questions
How many cyber resilience metrics should a security team report?
Track up to eight and report three or four. NIST SP 800-55 Vol. 1 warns that poorly selected quantitative metrics "can undermine the overall quality of reporting." The working set belongs to the people who own each row. The board set should stay the same quarter to quarter, because a metric that changes every quarter never becomes a trend.
Are there industry benchmarks for cyber resilience metrics?
I know of no credible benchmark for the human-response metrics in this article, and I would be wary of one that claims otherwise. Definitions differ on where each clock starts, samples are self-selected, and MITRE notes that comparison across organizations is hard to make meaningful. Compare against your own baseline, where the formula at least stayed the same.
What is the difference between a metric and a measure of effectiveness?
A metric is a value you collect, such as escalation latency in one exercise. It becomes a measure of effectiveness when you use it to judge a specific change, such as a new escalation path, by comparing observations before and after that change. One run gives you a metric. Runs on either side of a change give you an MOE.
Can cyber resilience metrics be collected without running exercises?
Yes. Real incidents produce every metric in this article, and a scribe working from a shared log can capture declaration, escalation, and decisions as they happen. The limit is volume. Most organizations face few serious incidents a year, so a series built only on real events fills slowly. Exercises add observations on a schedule you choose.
How often should cyber resilience metrics be reviewed?
Review the working set after every exercise or significant incident, and report the board set quarterly. NIST advises choosing collection frequency "based on a rate of change that is being evaluated." Decision latency moves with training and staffing, so quarterly suits it. Restore-order accuracy moves only when the recovery plan changes, so annual review may be enough.
Does NIST publish guidance on security measurement?
Yes. NIST SP 800-55 comes in two volumes, both dated December 2024. Volume 1 covers identifying and selecting measures, including measure types and a template for each one. Volume 2 covers building a measurement program. Neither defines cyber resilience metrics as such, so the definitions here are adaptations, not NIST text.
References
Bodeau, D., Graubart, R., McQuaid, R., & Woodill, J., Jr. (2018). Cyber resiliency metrics, measures of effectiveness, and scoring (MITRE Public Release No. 18-2579). The MITRE Corporation. https://www.mitre.org/news-insights/publication/cyber-resiliency-metrics-measures-effectiveness-and-scoring
Goldschmidt, C. (2026, October). Same incident, two teams: What Hacker Halted revealed about incident readiness. Reflex Security. https://reflexsecurity.io/blog/same-incident-two-teams-hacker-halted-incident-readiness
Schroeder, K., Trinh, H., & Pillitteri, V. (2024). Measurement guide for information security: Volume 1 — Identifying and selecting measures (NIST Special Publication 800-55 Vol. 1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.SP.800-55v1
Schroeder, K., Trinh, H., & Pillitteri, V. (2024). Measurement guide for information security: Volume 2 — Developing an information security measurement program (NIST Special Publication 800-55 Vol. 2). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.SP.800-55v2
