top of page

You Cannot Protect What You Cannot Map: Why Operational Resilience Is Making Dynamic Service Mapping Essential

Oct 26
13 min read

Updated: Nov 10

Frontier AI is accelerating vulnerability discovery and exploitation, while the FCA, PRA and DORA expect firms to show a detailed, current understanding of the dependencies behind their critical services. In this article we look at why static resilience maps are becoming harder to defend, how dynamic service mapping provides the technology dependency layer, where automation and AI help or mislead, and why the real goal is better decisions rather than CMDB accuracy.


Christian McCarron | Senior Technical Architect | Automiq

Author:

Christian McCarron

Senior Technical Architect



The emergence of frontier AI cyber capabilities such as Mythos changes an already difficult equation for large organisations. As we explored in an earlier post on Mythos, vulnerability discovery and exploitation are becoming faster, more automated and more scalable, and simply knowing that a vulnerability exists is no longer enough. Organisations require a clear plan to remediate or mitigate the vulnerability with an urgency proportionate to the threat – a topic we explored in our article on patching vulnerabilities in the age of Frontier AI.


Any plan, however, requires some difficult questions be answered before action can be taken: what sits behind that vulnerability, where the affected component is, what depends on it, which business services it supports, what an attacker could potentially reach from it, and what the operational consequences would be if it had to be isolated quickly.


For many organisations, the quality of those answers still depends on something considerably less fashionable than AI: the quality of the CMDB. That has always mattered for IT operations, but increasingly it matters for cyber security and Operational Resilience as well. The FCA, PRA and DORA are all pushing firms towards a clearer understanding of the dependencies that sit beneath critical or important business services. In practical terms, that means knowing not only which systems exist, but how they relate to one another and which business services depend upon them – a topic we explored in depth in our article on service mapping.


This is where service mapping becomes much more than an ITOM capability.


Dynamic Service Mapping for Regulatory Compliance | Automiq

Most large organisations can produce some form of asset inventory. They will usually know how many servers they have, which applications are in use and where at least the majority of their cloud resources are running. What is often much harder is understanding how all of those components work together. A vulnerable Linux package on a server is one piece of information; knowing that the server supports an integration tier, which feeds a payments application, which in turn underpins an Important Business Service with a two-hour impact tolerance is something very different.


That context changes the way a security team can respond. Instead of being told that 2,000 assets are affected by a vulnerability, they can distinguish between systems that are peripheral and those that sit directly in the delivery path of critical services. This becomes even more important as AI-driven security tooling improves. Better vulnerability discovery is undoubtedly useful, but it may also result in more findings than teams can realistically remediate at once, which makes prioritisation more important rather than less. A CVSS score can tell you how severe a technical weakness is, but it cannot tell you whether the vulnerable component is part of a customer-facing service, a back-office function, a recovery path or something genuinely non-critical. That context comes from understanding the service.


Operational Resilience is raising the standard for mapping


The regulatory direction makes this particularly relevant for financial services. Under the FCA's Operational Resilience framework, firms are expected to identify and document the people, processes, technology, facilities and information needed to deliver their Important Business Services, including relevant third-party dependencies. By 31 March 2025, firms in scope were expected to have completed the mapping and testing necessary to demonstrate that those services could remain within their impact tolerances.


The focus has therefore moved on. In 2026, the question is increasingly not whether an Operational Resilience framework exists, but whether firms can demonstrate that it is embedded and that their understanding of dependencies is sufficiently detailed to support vulnerability identification, testing and decision-making. The FCA's 2026 observations reinforce this direction, highlighting the importance of comprehensive mapping, multiple data sources, clear ownership and the use of mapping information to identify vulnerabilities and guide testing.


That represents a meaningful change from the way resilience mapping has often been approached historically. In many organisations a workshop is held, an architecture diagram is created, dependencies are captured in a spreadsheet and the result is signed off. At the point it is produced the map may be accurate, but the problem is what happens afterwards. Cloud resources change, firewall rules are amended, new integrations are introduced, containers are redeployed, applications are patched and third-party services evolve. The document is still there, and may still look authoritative, but the environment underneath it has moved on.

This is a problem I have seen repeatedly in large environments.

It is rarely the case that nobody has ever documented the service. More often, different teams hold different versions of the truth, produced at different times and for different purposes. The resilience team has one representation, IT Operations has another, Cyber Security has its own asset view and Enterprise Architecture has something else again. Each may be broadly right, but collectively they can still leave the organisation uncertain about which version should be trusted when a real incident occurs.


That is why the regulatory direction naturally points towards more dynamic forms of service mapping. The FCA does not mandate ServiceNow, a CMDB or any particular discovery platform, and it does not say that Visio diagrams or spreadsheets are unacceptable. However, there is an important difference between something being permitted and it being sustainable. If a firm is expected to understand vulnerabilities, dependencies, third parties and impact tolerances on an ongoing basis, then static mapping becomes progressively harder to defend in a large and changing technology estate.


Service mapping provides the technology dependency layer


At its simplest, service mapping connects technical components to the applications and business services they collectively deliver. Rather than merely recording that a database, Kubernetes cluster, VM or load balancer exists, the map shows how those components participate in the delivery of a service.


A mature model might trace a path from an Important Business Service through the business application and application service, down to application components, infrastructure, network and external dependencies. That kind of model has obvious operational uses. If a component needs to be isolated during a cyber incident, responders can understand the likely impact before taking action. If a new vulnerability is identified, teams can see which services inherit the exposure. If an outage occurs, support teams can understand the potential blast radius. If an Operational Resilience team is designing a scenario test, the same relationships can help identify genuine concentration risks, single points of failure and dependencies worth testing.


One of the strongest arguments for good service mapping is that the same underlying dependency model can support multiple disciplines. Operations, Cyber Security, Architecture, Business Continuity, Risk and Operational Resilience are often trying to answer closely related questions, yet many organisations still maintain separate artefacts for each of them. That duplication not only creates unnecessary effort, it creates inconsistency.


One pattern I have seen is the creation of a dedicated “OpRes map” that sits separately from operational technology data. It is understandable why this happens: resilience teams need something quickly, so they run workshops, speak to service owners, document important suppliers and applications, and create a regulatory view of the service. At the same time, IT Operations has a CMDB, Cyber Security has its own inventory, the network team has topology data and Business Continuity has another set of dependency records. Each may be defensible in isolation, but they are often not identical.


The more useful target is not to force every piece of resilience data into one platform, but to build a federated model in which the technology dependency layer is drawn from sources that genuinely understand the live environment. Other authoritative sources can then contribute information about people, processes, facilities, suppliers and business ownership. The important point is that the technology map should ideally be derived from operational evidence rather than recreated periodically for a compliance exercise.


ServiceNow is one implementation, not the definition


ServiceNow is one of the more mature platforms for implementing this approach, particularly where an organisation already uses Discovery, CMDB and Service Mapping. Traditional ServiceNow Service Mapping starts from an application entry point and uses discovery patterns to identify components and relationships, but no single discovery technique is sufficient in a modern enterprise.


Large estates now span physical servers, virtual machines, public cloud, containers, SaaS, managed services and third-party APIs. Useful dependency evidence may therefore come from discovery patterns, cloud APIs, application telemetry, load balancers, firewall data, DNS, Infrastructure-as-Code, observability platforms and network traffic.


Traffic-based discovery is particularly useful because it can reveal what systems are actually communicating with one another rather than simply what the architecture says should be happening. However, this is also where automation can become misleading. In practice, one of the hardest parts of service mapping is not discovering relationships, but deciding which relationships actually matter.


A backup product may communicate with hundreds of systems. A vulnerability scanner may touch large parts of the estate. Monitoring tools, administrative connections and temporary support sessions all create real network traffic, but they are not necessarily dependencies required to deliver the service. If every observed connection is automatically treated as a meaningful relationship, the map can quickly become too noisy to be useful.

This is why I am cautious about measuring CMDB maturity by the percentage of data that is automatically discovered.

A map containing thousands of relationships is not necessarily a good map simply because those relationships were generated by tooling. The real measure is whether the relationships are accurate enough to support operational and risk decisions.

The strongest model therefore combines automation with human validation. Discovery can identify infrastructure, traffic analysis can identify communication, observability can show behaviour and AI can identify anomalies or probable missing relationships. Application owners, service owners, architects and operational SMEs still need to validate whether those relationships genuinely form part of the service.


That validation matters technically, but it also matters from a governance perspective. There is a significant difference between saying that a tool discovered a relationship and saying that the relationship was identified from operational evidence, reviewed by the people who understand the service and accepted as part of the approved dependency model. For an Operational Resilience programme, the second position is much stronger because it provides ownership and accountability rather than relying on automation alone.


DORA is driving the same requirement from another direction


DORA creates similar pressure for EU financial entities. Its requirements extend beyond identifying ICT assets and into understanding which business functions those assets support, their criticality and the interdependencies between them. Third-party risk is also a major part of the framework.


Again, the challenge is not simply knowing that a third party exists. Knowing that an organisation uses AWS, Azure or a particular SaaS provider tells you relatively little about resilience on its own. The more useful questions are which services depend on that provider, whether alternative paths exist, whether several important services rely on the same underlying component, and whether one failure could create concentration risk across the organisation.


Those are dependency questions, and service mapping is fundamentally a way of answering them.


The FCA/PRA framework and DORA use different terminology and operate through different regulatory structures, but both are pushing firms towards a deeper and more defensible understanding of the dependencies supporting critical services. For UK firms in particular, the timing matters. The March 2025 deadline has passed, and the discussion is moving from framework design to supervisory scrutiny. The FCA's 2026 observations point towards more detailed mapping, stronger ownership, multiple sources of evidence and a better understanding of third-party dependencies. The PRA has similarly focused on improving resilience testing and embedding Operational Resilience into decision-making.


For firms that still rely heavily on periodic workshops, Visio diagrams and spreadsheets, that creates an immediate question: can those artefacts really provide a sufficiently current picture of a complex technology environment? In some smaller or relatively static organisations they may be adequate. In a large bank, insurer or similarly distributed enterprise, that becomes increasingly difficult.


Static documentation will not disappear. Architecture diagrams, resilience assessments and governance documents still have a role. The difference is that they should increasingly be views or outputs from a living dependency model, rather than the primary repository of knowledge.


AI can improve the mapping process, but it also introduces new risks


AI has obvious potential in this area because large organisations generate far more dependency information than people can realistically correlate manually. It can help analyse discovery data, network flows, cloud metadata, configuration records, application telemetry, DNS, change records, incident history and existing CMDB relationships. Used well, it can identify missing relationships, stale CIs, unusual communication patterns and areas where observed behaviour differs from the approved service model.


It also has the potential to make dependency data far easier to use. A security analyst should be able to ask which Important Business Services might be affected by a newly discovered vulnerability. A resilience practitioner should be able to ask which critical services share a particular third-party dependency. The underlying relationships may already exist in the CMDB; AI simply makes them easier to interrogate.


The risk is that AI is very good at producing plausible answers, and plausibility is not the same as correctness. A model may infer that two systems are dependent because they communicate frequently, even though the traffic relates only to monitoring or administration. Conversely, it may understate the importance of a component that sees very little traffic but provides a critical disaster-recovery function.

For that reason, I do not think AI should become the authority for service mapping.

A more sensible operating model is that AI proposes, discovery provides evidence, owners validate and governance approves. AI can reduce the manual effort involved in maintaining the map, but it should not remove accountability for deciding whether the map is correct.


The real goal is not CMDB accuracy


One of the biggest mistakes in CMDB programmes is treating “CMDB accuracy” as the objective in itself. The real objective is to enable better decisions.


Cyber Security and Operational Resilience increasingly need answers to the same questions.


  • What is vulnerable?

  • What does it support?

  • What can reach it?

  • What happens if it fails?

  • What happens if we isolate it?

  • Which Important Business Services depend on it?

  • Where are the single points of failure?

  • Which third parties are involved?

  • Can the service remain within impact tolerance, and who owns the risk?


If answering those questions requires several days of spreadsheet reconciliation, architecture workshops and conversations with the people who “know how it really works”, then the organisation does not have a mature mapping capability. It has institutional memory.


The opportunity created by Operational Resilience regulation is to move beyond that. Rather than treating FCA, PRA or DORA mapping as another compliance documentation exercise, organisations can use those requirements as the catalyst for building something more useful: a dynamic, evidence-based and human-validated model of how their business actually works.


That matters for resilience and for cyber risk. As capabilities such as Mythos make vulnerability discovery and exploitation faster, it becomes increasingly important that defenders understand their own technology estate better than an attacker does.


What enterprises should do now - and how Automiq can help



Frequently asked questions


What is dynamic service mapping?

Dynamic service mapping is the practice of continuously maintaining an accurate model of how technology components – servers, databases, cloud resources, containers, network devices and third-party services – support specific applications and business services. Unlike a one-off diagram, the map is derived from operational evidence such as discovery, cloud APIs, traffic analysis and observability data, then validated by service owners, so it stays current as the environment changes.


How is dynamic service mapping different from static documentation such as Visio diagrams or spreadsheets?

Static documentation captures dependencies at a single point in time, usually through workshops and interviews. It can be accurate when produced but quickly drifts as cloud resources, integrations, firewall rules and suppliers change. Dynamic service mapping keeps the dependency model aligned to the live environment, with static diagrams and resilience documents becoming views or outputs of that model rather than the primary source of knowledge.


Does the FCA require firms to use a CMDB or ServiceNow for Operational Resilience mapping?

No. The FCA does not mandate ServiceNow, a CMDB or any specific discovery platform, and it does not prohibit spreadsheets or diagrams. However, firms are expected to understand the people, processes, technology, facilities, information and third parties behind their Important Business Services on an ongoing basis. In large, fast-changing technology estates, maintaining that understanding manually becomes increasingly difficult to sustain and defend.


What did the FCA and PRA expect firms to have in place by 31 March 2025?

By 31 March 2025, firms in scope of the UK Operational Resilience rules were expected to have completed the mapping and scenario testing needed to show that their Important Business Services could remain within impact tolerances during severe but plausible disruption. The focus has since moved to demonstrating that Operational Resilience is embedded, and that mapping is detailed and reliable enough to identify vulnerabilities and inform decisions.


How does DORA relate to service mapping?

The EU's Digital Operational Resilience Act (DORA) requires financial entities to identify their ICT assets, understand which business functions they support, assess their criticality and map the interdependencies between them. It also places significant emphasis on ICT third-party risk and concentration risk. These are fundamentally dependency questions, which is why dynamic service mapping is a practical foundation for meeting DORA's expectations.


How does service mapping help prioritise vulnerabilities?

A CVSS score describes how severe a technical weakness is, but not how much it matters to the business. Service mapping adds that context by showing whether an affected component sits in the delivery path of an Important Business Service, a back-office function, a recovery path or something non-critical. That allows security teams to prioritise remediation by business impact – increasingly important as AI-driven tooling produces more findings than teams can fix at once.


Why is the percentage of automatically discovered CMDB data a poor measure of maturity?

Automated discovery can generate thousands of relationships, but many reflect backup, monitoring, vulnerability scanning or administrative traffic rather than genuine service dependencies. A map full of noisy relationships can be less useful than a smaller, accurate one. A better measure is whether relationships are accurate, owned and validated enough to support real operational, cyber and risk decisions.


Can AI build and maintain a service map on its own?

AI can significantly reduce the effort of maintaining a service map by analysing discovery data, network flows, telemetry, change records and incident history to propose missing relationships and flag stale data. It can also make dependency data easier to query. But AI produces plausible answers rather than guaranteed correct ones, so it should not be the authority. A sound model is that AI proposes, discovery provides evidence, owners validate and governance approves.


What is a federated dependency model for Operational Resilience?

A federated dependency model avoids forcing all resilience data into a single platform. Instead, the technology dependency layer is drawn from sources that understand the live environment, such as discovery and observability tools, while other authoritative sources contribute information on people, processes, facilities, suppliers and business ownership. This reduces duplication between Operations, Cyber Security, Architecture, Business Continuity and Risk, and gives everyone a consistent view.


Where should an organisation start with dynamic service mapping?

Start with a small number of Important Business Services that have the tightest impact tolerances. Reconcile the existing resilience maps, CMDB data, security inventories and architecture artefacts for those services, establish evidence-based discovery for the underlying technology, and assign named owners to validate the result. Then connect the map to vulnerability, incident and change processes so it informs decisions rather than sitting as documentation.

Automiq logo light
bottom of page