Explore our new Cybersecurity Terms Glossary, explained by our experts.
Cybersecurity Terms Glossary
A correlation rule links events from different sources or different times so that a pattern becomes visible. Individually each event is unremarkable; together they describe an attack.

What Is a Correlation Rule?
A correlation rule is detection logic that considers more than one event. Rather than alerting on a single log entry, it looks for a relationship: the same user failing to authenticate 40 times and then succeeding, a firewall denial followed by an outbound connection to the same address from a different host, or a privileged group change that no corresponding change request explains. The events are often individually harmless, which is exactly why the rule exists.
Platform vocabulary differs but the mechanics do not. IBM QRadar composes rules from building blocks and tests over a sliding window. Splunk Enterprise Security runs scheduled correlation searches that produce notable events. Microsoft Sentinel uses analytics rules with entity mapping, including near-real-time variants. Elastic Security offers an event correlation query language for sequences. In each case the engine must hold state between events, which is the source of most of the operational difficulty.
Two classic examples are worth building early. The first is impossible travel combined with a privilege-sensitive action, which joins identity provider logs against a reference list of high-value operations. The second is mass file access followed by a large outbound transfer, which catches data staging ahead of exfiltration and depends on your file audit logs actually being enabled.
Stateful logic is fragile in ways that single-event rules are not. Time windows assume synchronised clocks, so an unmanaged NTP configuration quietly disables an entire class of detections. Joins assume that a user is identified the same way in every source, when in practice one log holds jsmith, another DOMAIN\jsmith, and a third [email protected]; without consistent normalisation the rule matches nothing and fails silently. Delayed ingestion from one source means the second event arrives after the window has closed. And memory consumption grows with window length, which is why a 24-hour correlation window is usually a performance decision rather than a detection decision.
The hardest failure to notice is the silent one. A single-event rule that stops working produces no alerts and someone eventually asks why. A correlation rule that stops working also produces no alerts, and because it never produced many, nobody asks. Health monitoring that checks whether each rule has fired within an expected period is the only reliable defence.
Chained correlation is brittle because attackers do not follow the sequence you wrote. The alternative that has gained ground over the last few years is risk-based alerting: each suspicious behaviour contributes a score to the user or host it concerns, and a case is raised only when an entity accumulates enough risk from enough distinct techniques within a window. It tolerates missing steps and out-of-order events, which rigid sequences do not, and teams that adopt it commonly report large reductions in case volume while surfacing chains that no individual rule would have escalated.
The trade-off is that attribution must be reliable, so entity resolution across sources becomes the critical dependency rather than the rule logic itself. Most mature estates end up with both: a small number of high-confidence sequence rules for well-understood attacks, and a risk model underneath catching everything that does not fit a template. Both belong in the same rule lifecycle, with version control, tests, and an owner. At Privia Security, we build and maintain correlation content as part of our managed security services.
Every correlation rule is a detection rule, but not every detection rule correlates. A simple detection fires on a single event matching a pattern, such as a specific command line. A correlation rule needs relationships between multiple events, which means the engine must hold state across time. That requirement makes correlation rules more powerful and considerably more fragile to operate.
Long enough to contain the behaviour and short enough that the engine can hold the state. Brute force sequences work in 5 to 15 minutes; lateral movement patterns may need hours. Every additional hour costs memory and increases the chance of unrelated events coinciding. Test with historical data to find the shortest window that still captures the real activity you are targeting.
Because their normal output is very little, so a drop to zero looks like a quiet period. Common causes are clock drift between sources, a field rename after a platform upgrade, inconsistent username formats breaking a join, and one source falling behind on ingestion. Monitor the last-fired timestamp of every rule and alert when a rule has produced nothing for longer than expected.
It is more resilient for multi-stage activity, because it does not require events to arrive in a specific order or for every step to be observed. It is not a complete replacement: well-understood attacks with a reliable signature are still better served by a precise sequence rule that produces an immediately actionable alert. Most mature teams run both models side by side.
You May Be Interested In These
We use a few cookies that are needed to make this website work, and they are always on. We would also like to set analytics cookies, to see which pages are useful; functional cookies, so that our live chat works; and advertising cookies, to measure our campaigns. We will not set any of those until you say yes, and you can change your mind at any time from the Cookie Settings link in the footer.
Choose the cookies you are happy for us to use. Strictly necessary cookies are always on; everything else is off until you turn it on.
These cookies keep the website working: they hold your security checks, your form submissions and the choice you make here. The site cannot run without them, so they are always on.
These let our live chat open and remember your conversation, and let the newsletter sign-up work. Switch them off and the chat window will not appear.
These tell us which pages are read and how people find us, so that we can improve the site. The reports are about groups of visitors, never about you personally.
These let us measure our advertising and show you our adverts on other websites, such as LinkedIn, Google and Meta. Switch them off and you will still see adverts, but they will be less relevant.