This sheet is past its review date. Deadlines or advice may have changed since 5 Nov 2026.
Security metrics: which few numbers show the board whether you're getting safer?
This quarter, pick about seven measures tied to how attacks start, count each one against your full inventory and have management agree a target and tolerance for each. Then report them on one page every quarter with trend, target and the decision you need, and drop any measure that hasn't changed a decision in a year.
If you only do five things
Count outcomes, not activity.
Report what an attacker would find. Microsoft says reporting should shift from patches deployed to exposure reduced, detection coverage increased and time to mitigate compressed1.
Count against the whole inventory.
Give every share its denominator, including assets found this quarter. A system nobody listed never shows up as uncovered.
Agree targets and tolerances with management.
The UK Cyber Governance Code asks boards to require reporting at least quarterly and agree a tolerance for each metric, based on the agreed risk appetite4.
Build the set in this order
Which ways in matter most to you?
Start from your incidents, near misses and test findings of the last two years, then check the year's M-Trends and DBIR.
What would show each path closing?
Write each measure's formula, target, data source, owner and frequency, as NIST's suggested fields do3. Prefer ages, times and shares to counts.
What is it counted against?
Name the inventory behind each share. List hypervisors, appliances and OT that can't run an agent as gaps, not as out of scope.
What does good look like?
After three months of data, propose a target and tolerance per measure, in days, euros or share of critical services, and have management sign them off.
Is each measure still earning its place?
Review the set at planned intervals and after significant incidents or changes, as Regulation 2024/2690 requires of digital providers7. Drop any measure that changed no decision in a year.
A starter set, tied to how attacks start
Report each as a trend against its target and coverage. The linked sheets define the measures.
- Internet exposure. High-risk services open to the internet without an expiring exception, and assets others found first, from the attack surface sheet.
- Age of exploited vulnerabilities. Days the oldest known-exploited flaw has been open, and first-tier deadlines met, from the patching sheet. Only 26% of KEV flaws in Verizon's data were fully fixed in 2025, and full fixes took a median of 43 days2.
- Phishing-resistant MFA coverage. Accounts, admins first, that Conditional Access holds to the passkeys sheet's phishing-resistant methods. Entra's passwordless count includes phone sign-in8, which isn't phishing-resistant9.
- Standing privileged accounts. Tier 0 rights held permanently, not on activation, against the privileged access sheet's limits. Microsoft ranks tiered administration among the best safeguards1.
- EDR coverage. Devices and servers with EDR and tamper protection, as the endpoints sheet sets them up, out of all discovered; Defender's inventory lists devices not onboarded10. Count hypervisors apart: virtualization has traditionally sat outside EDR's scope11.
- Measured restore time. Hours to restore each critical service in its last test, against its target, from the backups sheet. A target not met in a test is unmet.
- Time to detect and contain. In real incidents and planted tests, with alerts proved monthly from the logging sheet and staff report speed from the awareness sheet. Mandiant's median dwell time rose to 14 days in 2025, from 1111.
What you can skip
- Counts of blocked attacks. They follow attacker volume and filter settings, not your risk.
- Phishing click rates as a target. A poorly designed test can show a better rate while saying less, NIST notes3. Report reports instead.
- Colors and maturity levels alone. NIST doesn't count color scales or rankings as measures3. Put a number behind each.
- Live dashboards for the board. It needs a quarterly page of trends and decisions, not live data.
Mistakes that cost the most
- Reports that ask for nothing. Management can't oversee what it's never asked to decide.
- Shares without a denominator. 98% EDR coverage of known devices can hide every hypervisor and appliance that can't run an agent.
- New definitions every quarter. Each change breaks the trend. Restate past values when you must.
- A green report in the quarter of a breach. The set measured the wrong things. Rework it after each significant incident.
What goes on the board page
- One row per measure. Value, four quarters of trend, target and tolerance, and why it moved; the NCSC tells boards to expect agreed target ranges12.
- Top risks and accepted risks. The top risks with the status of their measures6, and each accepted risk with its expiry date.
- Incidents and near misses. What happened, what it cost and what changed. A board may need only the main lessons, the NCSC says12.
- Decisions needed. Each with options, cost and a recommendation.
- The same layout every quarter. Consistent metrics are what show long-term trends, NIST notes3.
What management must do
- NIS2. Management approves the measures, oversees them and can be held liable; the measures include assessing their effectiveness13.
- Germany. Under BSIG § 38, management implements and oversees the § 30 measures, including effectiveness checks, and is liable to the entity14. The BSI's example answer adds KPIs, KRIs and a yearly audit15.
- Digital providers. Indicators for the security policy, one person reporting directly to management, and measurement results reported to it7.
- DORA. The management body approves the ICT risk framework and risk tolerance; the strategy sets KPIs and key risk metrics. It hears of major incidents and yearly lessons learned5. BSIG § 38 doesn't apply14.
Keep the numbers honest
- Reward reporting. Valuing the absence of reported problems teaches people to keep quiet, the NCSC warns12. Count near misses as a success; NIS2 lets entities report them voluntarily13.
- Show the worst case. Put the oldest exploited flaw and the slowest restore next to each median.
- Separate inventory growth. Report newly found assets on their own line, so a better inventory doesn't look like decline.
- Have someone else recompute it. Once a year, let someone outside the team rebuild two measures from raw data.
What changed in 2026
- The BSI recommends a regular report. Its October 2026 curriculum lists a resolution on a regular cyber risk report among the decisions to take, ideally within 2 to 6 weeks of management's first training6.
- Voice phishing overtook email. It opened 11% of Mandiant's 2025 intrusions, after exploits; email phishing fell to 6%11. Measure how the help desk handles reset calls.
- Microsoft discounted alert counts. Its 2026 report says they no longer show SOC maturity; measure detection fidelity and time to contain1.
What this looks like at your size
Small team, no security staff
- Ask your IT provider for four numbers a quarter: admins on phishing-resistant MFA, EDR coverage, oldest open exploited flaw, last restore time.
- Give the managing directors one page and file their decisions with the training record.