Module 02 · Lesson 12

Designing a detection pipeline that doesn't drown you

You can build perfect defenses and still get breached. The difference between a minor incident and a catastrophe is how quickly you detect it.

200+
Days avg dwell time
10K+
Alerts/day avg SOC
95%
False positive rate
What to Log

Log what answers three questions

Must Log
Security-relevant events
Auth events (success + failure), authorization decisions, privileged actions, data access, network connections, system changes, security tool events.
Never Log
Creates new risk
Passwords (even failed — often 1 char off), full credit card numbers, PII in plaintext, health data. Use reference IDs instead.

Three questions your logs must answer: Did something bad happen? What exactly? How do we stop it?

SIEM Architecture

The nervous system of security ops

ComponentFunctionKey consideration
CollectionAgents, syslog, APIs pulling from sourcesCoverage: missing one server = blind spot
NormalizationConvert different formats to common schemaWithout it, can't correlate Windows + Linux events
CorrelationConnect related events across time/sourcesLogin from Madrid, then Beijing 5 min later = alert
DetectionRules that trigger alerts on patternsStart with vendor rules, then customize heavily
DashboardsVisibility for analysts and executivesSOC: real-time detail. CISO: weekly trends.
RetentionHow long logs are searchable/archivedHot: 30-90 days. Cold: 12-24 months.
Vendor Landscape

Choosing a SIEM

Splunk
Most powerful, most expensive. Dominant in enterprise. Volume-based pricing can surprise you.
Sentinel
Microsoft's cloud SIEM. Great for M365/Azure shops. Consumption pricing — monitor costs carefully.
Elastic
Open source option. Powerful but requires expertise to operate. Self-hosted or cloud.
Chronicle
Google's SIEM. Fixed pricing model — predictable costs regardless of volume.
Wazuh
Open source, free. Best option for SMBs starting from zero. Endpoint + SIEM in one.
Alert Fatigue

The silent killer

10,000+ alerts/day. Most false positives. Your analysts miss the real attack on page 47 of the queue. This is how breaches go undetected despite having a SIEM.

Tune
Fix or disable noisy rules. 50 false positives/week? Fix the rule. 20 well-tuned rules beat 500 noisy ones.
Tier
Not everything is critical. Critical (immediate), High (1hr), Medium (4hr), Low (batch review).
Automate
SOAR for the repetitive. Auto-enrich IPs, auto-block known-bad indicators, auto-close whitelist matches.
Measure
Track: false positive rate, MTTD, MTTR, alert-to-investigation ratio. If 95% are false positives, your SIEM is a liability.
Detection Maturity

Four levels of detection capability

Level 1
Basic Alerts
Vendor-provided rules, minimal tuning. High noise. Reactive only. Where most SMBs start.
Level 2
Tuned Detection
Custom rules for your environment. False positives managed. Incident response tested. Most mid-market target.
Level 3
Proactive Hunting
Dedicated threat hunters. Hypothesis-driven searches. Behavioral analytics. MITRE ATT&CK mapped coverage.
Level 4
Automated Response
SOAR playbooks auto-contain threats. ML-driven anomaly detection. Continuous improvement loop. Enterprise target.
Key Takeaway

Signal over noise

Remember this

200+ day average dwell time means most organizations don't know they've been breached for months.

A SIEM with 20 well-tuned rules beats one with 500 noisy rules. Quality over quantity.

Alert fatigue kills detection capability. Tune relentlessly, tier alerts, automate the repetitive.

Log everything security-relevant. Never log passwords or PII in plaintext.

1 / 7