Chapter 29

SOAR Platforms and IR Automation

SOAR platforms automate the repetitive actions that slow down IR — enrichment lookups, ticket creation, notifications, and containment triggers. Used well, they let analysts focus on judgment calls instead of copy-paste work. Used poorly, they create automation debt and alert-to-action paths nobody trusts.

Scenario

The SOC is handling 200 alerts per day. 180 of them follow identical patterns: an alert fires, an analyst opens it, looks up the IP in VirusTotal, checks the user account in AD, looks up the asset in the CMDB, determines it's a known false positive or low-risk event, and closes the ticket — 8 minutes of mechanical work per alert. The other 20 alerts require actual investigation. The team is spending 80% of their time on the 90% of alerts that don't need human thought, and 20% of their time on the 10% that do. SOAR automation's job is to flip that ratio — or, better, to eliminate the mechanical work entirely so all analyst time goes to real investigation. This chapter explains how SOAR platforms work and what makes implementations succeed or fail.

What SOAR Is

SOAR stands for Security Orchestration, Automation, and Response. The three words describe three distinct functions that SOAR platforms combine. Understanding what each means helps clarify where automation is useful and where it's not.

  The Three Functions of SOAR
  ══════════════════════════════════════════════════════════════════════

  ORCHESTRATION — connecting tools
  ──────────────────────────────────
  SOAR sits between your security tools and talks to all of them:
    EDR ──────────────────────────────────────────┐
    SIEM ─────────────────────────────────────────┤
    Firewall ─────────────────────────────────────┤
    Threat intel feeds ───────────────────────────┤
    AD / Identity providers ──────────────────────┤── SOAR ── Analyst
    Ticketing system (Jira, ServiceNow) ──────────┤
    Email system ─────────────────────────────────┤
    Case management (TheHive, ServiceNow) ─────────┤
    Notification systems (Slack, PagerDuty) ───────┘

  Without SOAR: analyst opens 8 tabs, runs 8 queries manually
  With SOAR: one playbook call pulls all data into one view

  AUTOMATION — removing human action from low-judgment tasks
  ──────────────────────────────────────────────────────────
  The question for every manual action:
    "Does this require human judgment, or is it mechanical?"
    Mechanical → automate it
    Requires judgment → assist the human, don't replace them

  Types of automation:
    Enrichment: "This IP appeared in alert X.
                 Auto-query: VirusTotal, Shodan, GeoIP,
                 internal CMDB, open port scan"
    Notification: "Create Jira ticket, page on-call,
                   post to Slack #incidents"
    Containment: "Isolate host in EDR, block IP in firewall,
                  disable user account in AD"
    (Containment automation = highest risk, highest value,
     requires most careful design)

  RESPONSE — structured playbook execution
  ──────────────────────────────────────────
  Translates IR playbooks into executable workflows:
    Each playbook = a directed graph of steps
    Each step = action (automated) or decision (human)
    Automation handles the actions; humans make the decisions
    Every action and decision is logged → full audit trail

Three Categories of Automation, by Risk

Not all automation carries the same risk. The highest-value SOAR implementations automate aggressively in low-risk categories while applying careful human oversight in high-risk actions. Organizations that build automated containment before they've mastered enrichment automation are making an ordering mistake that will result in missed alerts, incidents, and organizational distrust of the platform.

Automation categoryExamplesRisk levelWhen to automate
Enrichment automationIP reputation lookup, domain WHOIS, user/asset context from CMDB, threat intel feed correlationLow — read-only, no system changesAlways — this is where to start. Every alert should arrive to an analyst pre-enriched. No analyst should manually query VirusTotal for every alert.
Notification and ticketingCreate Jira ticket, post alert details to Slack, page on-call analyst, send SITREP emailLow — no security system changesAlways — alert routing and notification should never be a manual step
Closing known false positivesAuto-close alert if source IP matches internal scanner, auto-close if user activity is in expected business hours from enrolled deviceMedium — risk of over-suppressing real alertsAfter 90+ days of data showing 99%+ accuracy of the suppression logic; with human review of auto-closed alerts at least weekly
Soft containmentBlock IP in external firewall (no lateral movement from this IP), redirect user to quarantine VLAN (still monitored), pause account login (alert analyst to confirm)Medium — reversible, limited business impactWhen containment action is reversible in <5 minutes and business impact of false positive is low (no production system, no executive account)
Hard containmentIsolate host from network (EDR isolation), disable user account, block domain at DNSHigh — can interrupt business operations; hard to reverse quicklyRequires human approval for most scenarios; full automation only for confirmed high-confidence indicators (e.g., active ransomware encryption detected by EDR)
The automated containment failure mode: blocking your own security team

Organizations that automate containment actions without adequate exclusion lists create scenarios where the automation trips on legitimate security operations. Common examples: automated IP blocking that blocks your IR firm's external IP when they start a penetration test; automated account disabling that fires on a security analyst using abnormal search patterns during an investigation; automated host isolation that isolates the SIEM server itself when it generates unusual network traffic during a large log import. These failures create operational chaos during the moments when you most need things to work. Before automating any containment action, build explicit exclusion lists for your own security tool infrastructure, your IR firm's IPs, known high-privilege security accounts, and any system where an outage would be immediately business-critical. Review the exclusion list quarterly.

Designing SOAR Playbooks

A SOAR playbook is an IR playbook translated into executable workflow. Good playbook design starts from the human playbook, not from the SOAR platform's capabilities. The most common implementation mistake is to design around what the platform can easily do rather than what the IR process requires.

  Phishing Triage Playbook — Design Pattern
  ══════════════════════════════════════════════════════════════════════

  TRIGGER: Phishing alert from email security gateway
           OR user report to phishing inbox
  ↓
  AUTOMATED ENRICHMENT (no human needed):
    ├── Extract: sender, sender domain, subject, all URLs, attachments
    ├── Check sender domain: WHOIS age, MX records, reputation score
    ├── URL analysis: VirusTotal, URLscan.io, custom sandbox detonation
    ├── Attachment analysis: hash lookup, sandbox detonation if unknown
    ├── Recipient lookup: who received this email? AD group membership?
    │   Is recipient on high-value target list (finance, executives)?
    └── Similar email search: did this email land in other mailboxes?
  ↓
  AUTOMATED DECISION (based on enrichment results):
    ├── All clean → auto-close, log result, notify user "not a threat"
    ├── Clearly malicious → proceed to containment
    └── Uncertain → escalate to analyst with full enrichment context
  ↓
  AUTOMATED CONTAINMENT (if confirmed malicious):
    ├── Quarantine all copies of the email across all mailboxes
    ├── Block sender domain at email gateway
    ├── Create case in TheHive with all enrichment data
    └── Notify affected recipients
  ↓
  HUMAN DECISION POINT:
    ├── Did any recipient click the link/open attachment?
    │   YES → initiate host investigation playbook
    ├── Is this part of a broader campaign?
    │   YES → escalate to campaign scope playbook
    └── Close case if no further action needed
  ↓
  AUTOMATED CLOSE:
    ├── Log final disposition
    ├── Update threat intel feed with IOCs
    └── Close ticket with full audit trail

SOAR Maturity Progression

Organizations that try to implement full SOAR capabilities from day one consistently fail — the platform becomes a complex tool that nobody trusts because it breaks workflows in subtle ways. The correct progression is incremental, starting with the highest-value, lowest-risk automation and building trust before adding automation with higher business impact.

  SOAR Maturity Model: 4 Stages
  ══════════════════════════════════════════════════════════════════════

  STAGE 1: Alert enrichment and routing (Month 1-3)
  ─────────────────────────────────────────────────
  Automate: all enrichment lookups, ticket creation, analyst routing
  Measure: time-to-enrichment, analyst satisfaction with context quality
  Success: analyst never opens a blank alert; all context is pre-populated
  DO NOT YET: any containment actions

  STAGE 2: Known false positive suppression (Month 3-6)
  ──────────────────────────────────────────────────────
  Automate: auto-close for specific alert+enrichment combinations with
            confirmed ≥99% accuracy
  Measure: auto-close accuracy (sample review weekly), time saved
  Success: 30-50% of low-fidelity alerts auto-closed; zero missed incidents
           from auto-closing in 6-month review
  DO NOT YET: hard containment

  STAGE 3: Soft containment automation (Month 6-12)
  ──────────────────────────────────────────────────
  Automate: reversible containment actions for specific high-confidence
            alert types (phishing email quarantine, external IP block)
  Measure: containment time improvement, false positive containment rate
  Success: mean time to contain decreases; false positive containment
           reversal is uncommon and follows documented process
  DO NOT YET: account disabling, host isolation on non-confirmed cases

  STAGE 4: Tiered automated containment (Month 12+)
  ──────────────────────────────────────────────────
  Automate: host isolation and account disabling for specific confirmed
            indicators (active EDR ransomware detection, confirmed C2 IOC)
  Measure: mean time to contain for covered alert types
  Success: sub-5-minute containment on ransomware alerts;
           documented false positive analysis for any automated
           containment actions that were incorrect

Major SOAR Platforms

The SOAR market has consolidated significantly since 2020, with most capability now embedded in larger security platforms rather than available as standalone products. Understanding the landscape helps you evaluate what you may already have versus what requires separate procurement.

PlatformPositioningStrengthsConsiderations
Splunk SOAR (formerly Phantom)Standalone SOAR, integrates with any SIEMLargest library of app integrations; Python-based custom actions; strong communityExpensive licensing; significant implementation effort; Splunk acquisition may affect roadmap
Microsoft Sentinel (with Logic Apps)SIEM + SOAR in one platform; Azure-nativeFree with Sentinel licensing; deep Microsoft 365/Azure integration; low-code automationLess flexible than dedicated SOAR; better for Microsoft-heavy environments
Palo Alto XSOAR (formerly Demisto)Enterprise SOAR; integrated with Cortex platformIncident war room; marketplace of 600+ integrations; Python/JavaScript automationHigh cost; best ROI in Palo Alto-heavy environments
IBM QRadar SOAREnterprise SOAR; integrated with QRadar SIEMStrong regulatory workflow support; privacy module for breach notification trackingComplex implementation; best in IBM ecosystem
n8n / open-source alternativesLow-code workflow automation (not security-specific)Free; flexible; good for smaller teams building custom enrichment workflowsNot purpose-built for security; requires more engineering work; limited out-of-box security integrations
Why SOAR implementations fail

The majority of SOAR implementations underdeliver. The most common reasons: (1) Technology-first design — the team starts by learning what the platform can do and building playbooks around its capabilities, rather than starting from the IR process and translating it into automation. The result is automation for automation's sake. (2) Insufficient process documentation — you cannot automate an undocumented process. If the IR team doesn't have written playbooks before the SOAR implementation, the SOAR project becomes the first time the process is ever documented, which is extremely expensive. (3) Neglecting exception handling — every automated playbook needs to handle the cases where the automation doesn't know what to do. SOAR playbooks that reach a dead end and silently stall are worse than no automation. (4) No measurement — if you don't measure mean time to respond before and after SOAR implementation, you cannot demonstrate value. This matters for budget justification and for identifying which playbooks are actually saving time. (5) Treating automation as permanent — attacker TTPs change, tooling changes, environments change. Playbooks that aren't regularly reviewed become stale and eventually harmful — they auto-close or auto-contain based on logic that no longer reflects reality.

Q & A

Q: An automated SOAR playbook isolated a critical production server based on an EDR alert. The alert was a false positive. The isolation caused a 2-hour production outage. How do you prevent this from happening again?

Address it at three levels: (1) Immediate fix — add the affected server to an explicit exclusion list in the SOAR playbook. Any critical production system should be tagged in the CMDB as "high-business-impact" and that tag should cause the playbook to route to human approval instead of automated action. (2) Process fix — review all automated containment playbooks for the complete inventory of systems they could act on. Build a comprehensive exclusion list that covers: all Tier-1 production systems, all security infrastructure (SIEM, EDR management, SOAR itself), all backup and recovery systems, all executive workstations (high visibility false positive = immediate escalation). The exclusion list should be maintained as a living document, reviewed monthly, and require formal approval to modify. (3) Monitoring fix — any automated containment action should trigger a human notification within 60 seconds. Not for approval — the action already happened — but for immediate awareness so a human can reverse it quickly if it's a false positive. The 2-hour outage was avoidable if someone had been notified at T+1 minute that the server was isolated and reviewed the alert immediately. Build a "SOAR took an automated containment action" alert that pages an analyst every time, regardless of what the action was.

Q: How do you measure whether your SOAR implementation is actually working?

Three primary metrics with specific measurement methodology: (1) Mean time to enrich (MTTE): measure from alert creation time to the time when enrichment data is available to the analyst. Before SOAR: measure how long analysts spend on enrichment manually. After SOAR: measure how long the playbook takes to run. Target: sub-2-minute enrichment for most alert types (automated lookups are fast; the bottleneck is usually sandbox detonation). (2) Analyst alert handling time: measure the total time an analyst spends on an alert from first view to close. This is the number that matters most to the team and to capacity planning. Before SOAR: baseline this by sampling analyst time. After SOAR: track via the case management system timestamps. A successful SOAR implementation should reduce routine alert handling time by 50–70%. (3) Auto-close accuracy: for alerts that are automatically closed without analyst review, sample weekly and verify disposition was correct. Track false negative rate — real incidents auto-closed. This number should be zero (or within tolerance defined by the CISO). If it's not zero, the auto-close logic has a gap that needs immediate attention. Secondary metrics: playbook success rate (what percentage of playbook executions complete without errors), coverage rate (what percentage of alert types have a playbook), and time savings in analyst hours per month.

Q: An analyst complains that the SOAR enrichment is often wrong or outdated — the VirusTotal score doesn't match their manual check, the CMDB has the wrong asset owner. How do you address data quality problems in SOAR enrichment?

SOAR automation is only as good as the data sources it queries. Data quality problems are the most common reason analysts stop trusting SOAR output: (1) CMDB accuracy is foundational. SOAR enrichment that queries an inaccurate CMDB produces wrong context. Address CMDB accuracy as a separate initiative — SOAR implementation often surfaces how bad the CMDB actually is. (2) VirusTotal score staleness: VT scores change over time. If the SOAR is caching VT results, set the cache TTL aggressively (hours, not days) for IP reputation data. IPs get recategorized quickly. For URL data, re-check at the time of the alert, not from a cached result. (3) Build analyst feedback into the workflow: when an analyst overrides SOAR enrichment with their own manual check and gets a different result, that should trigger a process to identify why the discrepancy occurred. Was the data source stale? Was the integration broken? Systematic review of overrides finds data quality gaps. (4) Show data source and timestamp with every enrichment result. "VirusTotal score: 2/73 as of 4 days ago" is very different from "VirusTotal score: 2/73 as of 10 minutes ago." Stale data presented without context is worse than no data — it creates false confidence. When analysts can see that a result is 4 days old, they know to manually re-check. When the timestamp is hidden, they may trust old data.