Most resilience dashboards report how many backups ran and how many plans were reviewed, which says almost nothing about whether an important service can be brought back inside the time the business and its regulators will tolerate. This guide sets out the recovery, continuity and impact tolerance measures that do answer that question: how long it takes to recover each important business service, how often tolerances are breached in tests and real incidents, what test results show, and what DORA and UK operational resilience rules expect firms to evidence. It is written for resilience leads, business continuity managers, CISOs and risk teams who need a short, defensible set of numbers for the board. It deliberately stays with recovery and continuity measures; how to design a security metrics programme as a whole is covered in the Security Metrics and KPIs Programme Guide.
Key takeaways
- Measure services, not systems. Resilience metrics are anchored on important business services and their impact tolerances; system RTOs only matter as inputs to the service figure.
- Impact tolerance is a time plus, where useful, a volume. The PRA requires every impact tolerance to include a time-based metric and allows additional metrics, such as the volume of interrupted transactions.
- Tolerance breaches are the headline KPI. Count them in severe but plausible scenario tests and in live incidents, and report the scenarios where you could not stay within tolerance, as the PRA expects in the self-assessment.
- Cyber recovery needs its own measures. Time to a clean recovery environment, time to restore identity services and verified backup integrity tell you more about a destructive attack than RTO and RPO alone.
- DORA turns several measures into obligations. Yearly testing of continuity and recovery plans, RTO and RPO per function, incident reporting clocks and an estimate of annual costs and losses from major incidents all produce numbers you should already be tracking.
The vocabulary behind the numbers
Several terms overlap, and dashboards often mix them. Keep them distinct so that each metric has one meaning. For the wider resilience picture these feed into, see the Cyber Resilience Guide.
| Term | Defined by | Applies to | What it is |
|---|---|---|---|
| Important business service | PRA SS1/21, FCA PS21/3 | Service the firm provides to external users | A service the firm identifies, with its rationale, and for which it must set an impact tolerance |
| Impact tolerance | PRA SS1/21 para 3.1 | Important business service | The maximum tolerable level of disruption, expressed with a time-based metric and, where appropriate, other metrics |
| Critical or important function | DORA | Function of an EU financial entity | The unit against which DORA sets RTOs, RPOs and yearly testing |
| Maximum tolerable downtime (MTD) | NIST SP 800-34 Rev 1 | Mission or business process | Total time a process can be disrupted before unacceptable consequences |
| Recovery time objective (RTO) | NIST SP 800-34 Rev 1; DORA Art. 12(6) | System or function | Target time to restore the system or function, which must fit inside the MTD or tolerance |
| Recovery point objective (RPO) | NIST SP 800-34 Rev 1; DORA Art. 12(6) | Data | The point in time to which data must be recoverable, which sets backup or replication frequency |
The relationship matters for measurement: system RTOs roll up, through dependencies, to a service recovery time, which is what you compare with the impact tolerance. If the slowest critical dependency has an RTO longer than the tolerance, the service cannot meet its tolerance however good the other components are.
Impact tolerance metrics
The PRA's SS1/21 says an impact tolerance must include a time-based metric, stating that a service should not be disrupted beyond a certain period or point in time, such as 24 hours or the end of the day, and that firms should add other metrics where appropriate. The FCA's rules for solo-regulated firms came into force on 31 March 2022, and both regulators required firms to be able to remain within tolerance by no later than 31 March 2025. These are the measures that show whether that position holds.
| KPI | Definition | Source of data | Good direction |
|---|---|---|---|
| Tolerance headroom | Impact tolerance minus demonstrated service recovery time, per service, from the latest test | Scenario test reports | Positive and stable or growing |
| Tolerance breaches in tests | Number of severe but plausible scenarios in which a service exceeded its tolerance, per period | Test programme | Falling, with each breach tied to a remediation action |
| Tolerance breaches in live incidents | Number of real disruptions that exceeded tolerance, with duration over tolerance | Incident records | Zero; any breach goes to the board |
| Secondary metric breaches | Breaches of non-time tolerance measures, for example transactions not processed or customers affected | Service telemetry | Zero |
| Services tested in period | Percentage of important business services tested against at least one severe but plausible scenario this year | Test plan | 100 percent |
| Open vulnerabilities to tolerance | Count and age of identified vulnerabilities that threaten a tolerance, against the justified completion date | Self-assessment action log | Falling; none past due |
| Mapping currency | Percentage of services whose dependency mapping was reviewed in the last 12 months or after significant change | Mapping repository | 100 percent |
Two design choices keep these honest. First, report breaches by scenario, not just a total, because the PRA expects firms to document the scenarios under which they could not remain within tolerance and to test scenarios where they expect to exceed it. Second, keep a record of lessons learned and actions taken, from tests and real incidents, because the self-assessment must contain them.
Recovery and continuity KPIs
Below the tolerance figures sit the operational measures that explain them. These are the numbers a resilience team watches monthly.
- Demonstrated RTO attainment: percentage of critical systems whose last full restore met its RTO. A system that has never been restored in full counts as not met.
- Demonstrated RPO attainment: percentage of critical data sets where the last test recovered data within the RPO. Measure the actual age of recovered data, not the scheduled backup frequency.
- RTO coverage gap: number of dependencies whose RTO is longer than the impact tolerance of a service that relies on them.
- Restore success rate: successful restore tests divided by attempted restore tests, by platform. See the Backup and Recovery Testing Checklist.
- Plan test currency: percentage of continuity and recovery plans tested in the last 12 months and after substantive change.
- Time to invoke: elapsed time from detection of a disruption to the decision to invoke continuity or recovery arrangements. Slow invocation consumes tolerance before recovery starts.
- Third-party recovery evidence: percentage of material suppliers supporting important services that have provided test evidence of their own recovery times.
Cyber recovery measures beyond RTO and RPO
RTO and RPO were designed for failures where the infrastructure is trustworthy. In a destructive cyber attack, the questions are different: is the backup clean, can identity be rebuilt, and how quickly can a trusted environment be stood up. These measures address them.
| Measure | Why it matters | How to measure |
|---|---|---|
| Time to clean recovery environment | Recovery cannot start until there is somewhere trusted to restore into | Elapsed time in an exercise from invocation to a usable isolated environment with tooling and network |
| Time to restore identity services | Directory and authentication services are a dependency of nearly every application | Time to recover the directory forest or tenant configuration from a known-good copy in a test |
| Verified backup integrity | NIST CSF 2.0 RC.RP-03 expects backup integrity to be verified before restoration | Percentage of critical backups scanned or validated as clean, and the date of last validation |
| Immutable or isolated copy coverage | Attackers target backup platforms and their credentials first | Percentage of critical data with an immutable or logically isolated copy under separate credentials |
| Clean point identification time | Restoring too recent a copy can reintroduce the attacker | Time in an exercise to agree the last known-good recovery point |
| Full service rebuild time | The real test of a ransomware scenario | End-to-end time to rebuild one important service from bare infrastructure in an exercise |
For the backup controls behind these figures, see the Backup Security and Immutable Backup Guide. The identity side matters as much: backup administrators, recovery service accounts and the keys that decrypt backups are non-human and privileged identities whose compromise defeats every recovery KPI. The Backup Identity Isolation Guide covers separating them.
DORA and impact tolerance reporting
Regulation (EU) 2022/2554, the Digital Operational Resilience Act, has applied since 17 January 2025. Several of its articles produce figures that a resilience metric set should hold ready. See the DORA Guide for the regulation as a whole.
| DORA provision | Requirement (summarised) | Metric to hold |
|---|---|---|
| Art. 11(6) | Test ICT business continuity plans and ICT response and recovery plans at least yearly and after substantive changes to systems supporting critical or important functions | Plan test currency; test outcomes against RTO |
| Art. 11(10) | Entities other than microenterprises report, on request, an estimate of aggregated annual costs and losses caused by major ICT-related incidents | Annual costs and losses per major incident, with method |
| Art. 12(6) | When setting recovery time and recovery point objectives for each function, consider whether it is critical or important and its potential market impact | RTO and RPO per function, and attainment in tests |
| Art. 13(2) and 13(5) | Post-incident reviews after major incidents that disrupt core activities; senior ICT staff report at least yearly to the management body on lessons learned | Reviews completed; actions closed on time |
| Art. 18(1) | Classify incidents by clients affected, duration including service downtime, geographical spread, data losses, criticality of services and economic impact | Downtime and clients affected per incident, captured at the time |
| Art. 19(4) with Delegated Regulation (EU) 2025/301 Art. 5 | Initial notification within four hours of classification as major and no later than 24 hours from awareness; intermediate report within 72 hours of the initial notification; final report no later than one month after the latest intermediate report | Reporting timeliness against each clock |
| Art. 24(6) and Art. 26(1) | Appropriate tests at least yearly on all systems supporting critical or important functions; TLPT at least every three years for entities required to perform it | Testing coverage; findings remediated; next TLPT date |
DORA's incident classification criteria and the UK impact tolerance regime overlap but are not the same: one sets reporting thresholds, the other sets the maximum disruption to a service. Capture duration, clients affected and data loss for every significant incident once, and derive both views from the same record. Testing design is covered in the DORA Resilience Testing Guide.
Building the metric set
- Fix the scope: list the important business services or critical and important functions, each with an owner and a tolerance or MTD agreed by the business, not IT.
- Map dependencies: for each service, list the systems, data, identity services, third parties and people it needs, with each dependency's RTO and RPO.
- Find the gaps on paper: flag any dependency whose RTO exceeds the service tolerance. This is the RTO coverage gap, and it is usually the first useful number.
- Test to produce evidence: run restore tests and severe but plausible scenarios, recording demonstrated recovery time per service. Include at least one destructive cyber scenario per year.
- Pick a short board set: three to six measures, typically tolerance breaches, headroom by service, plan test currency, cyber recovery rebuild time and overdue remediation.
- Automate collection: pull restore results from backup platforms, incident durations from the ticketing system and test outcomes from the exercise log, so figures are not assembled by hand each quarter.
- Review annually or after change: the PRA expects important business services and mapping to be reviewed at least annually or sooner after significant change; review the metrics on the same cycle.
Decision guidance
- Regulated UK financial firm: lead with impact tolerance metrics and self-assessment evidence; RTO and RPO are supporting detail.
- EU financial entity under DORA: lead with per-function RTO and RPO, yearly test results and incident reporting timeliness; add tolerance-style headroom if the board wants a service view.
- Unregulated organisation: borrow the impact tolerance idea anyway. A single time figure per critical service, agreed by its business owner, makes every other recovery number interpretable.
- Small organisation: track four figures: restore success rate, last full restore date for each critical system, time to restore identity and email, and number of plans tested this year.
- Large or group structure: keep one metric definition set for all entities and report at service level, so a shared platform outage shows up in every affected service rather than once in IT.
For how to position these measures in a wider maturity journey, see the Cyber Resilience Maturity Model.
Framework mapping
| Framework | Reference | Metrics it supports |
|---|---|---|
| NIST SP 800-53 Rev 5 | CP-2 Contingency Plan; CP-2(8) Identify Critical Assets | Service and dependency scope, mapping currency |
| NIST SP 800-53 Rev 5 | CP-4 Contingency Plan Testing; CP-9 System Backup; CP-10 System Recovery and Reconstitution | Plan test currency, restore success rate, RTO and RPO attainment |
| NIST SP 800-53 Rev 5 | IR-4 Incident Handling; PM-6 Measures of Performance | Time to invoke, live tolerance breaches, metric programme |
| NIST CSF 2.0 | GV.OC-04; RC.RP-01; RC.RP-03; RC.CO-03; ID.IM-02 | Critical services, recovery execution, backup integrity, recovery communication, improvements from tests |
| ISO/IEC 27001:2022 | Annex A 5.29 Information security during disruption; 5.30 ICT readiness for business continuity; 8.13 Information backup | Continuity and backup measures |
| UK and EU regulation | PRA SS1/21; FCA PS21/3; DORA Arts. 11 to 13, 18, 19, 24 and 26 | Impact tolerance, testing and reporting measures |
Common failure patterns
- Reporting activity as resilience: backup job success and plan review counts on the dashboard, but no demonstrated recovery time for any service.
- RTOs set by IT, tolerances by nobody: recovery targets chosen from what the infrastructure can do rather than what the business can tolerate, so every test passes and every incident surprises.
- Testing components, never services: each system restores within its RTO in isolation, but the end-to-end service, with identity, network and third parties, has never been recovered together.
- Assuming backups are clean: restore success measured, integrity not, so the first full restore after a ransomware event reintroduces the attacker.
- Losing incident data: duration, clients affected and data loss reconstructed weeks later from memory, which undermines both DORA classification and tolerance breach reporting.
- Hiding breaches in averages: a mean recovery time across services that looks healthy while one important service breaches its tolerance in every test.
Practitioner checklist
- Agree an impact tolerance or MTD with each important service owner, with a time metric and any secondary metric.
- Map dependencies per service and flag every dependency whose RTO exceeds the service tolerance.
- Record demonstrated recovery time per service from tests, and calculate headroom against tolerance.
- Count tolerance breaches separately for tests and live incidents, by scenario, with a remediation owner for each.
- Add cyber recovery measures: clean environment time, identity restore time, verified backup integrity and full rebuild time.
- For DORA entities, hold RTO and RPO per function, yearly test results, reporting timeliness and an annual costs and losses estimate.
- Capture duration, clients affected and data loss at the time of every significant incident.
- Report three to six measures to the board and review the set at least annually or after significant change.
Standards and references
- EUR-Lex: Regulation (EU) 2022/2554 on digital operational resilience for the financial sector (DORA)
- EUR-Lex: Commission Delegated Regulation (EU) 2025/301 on major ICT-related incident reporting
- Bank of England PRA: SS1/21 Operational resilience, impact tolerances for important business services
- FCA: PS21/3 Building operational resilience
- Basel Committee on Banking Supervision: Principles for operational resilience
- NIST: SP 800-34 Rev 1 Contingency Planning Guide for Federal Information Systems
Related NHI Mgmt Group resources: Cyber Resilience Guide · Security Metrics and KPIs Programme Guide · Cyber Resilience Maturity Model · Business Continuity and Disaster Recovery Guide