SOAR in the SOC: Which steps to automate and which still need an analyst?
SOAR in the SOC connects data, tools, and workflows to automatically execute predefined steps in security monitoring and incident response. SOAR can create cases, enrich data, route alerts, send notifications, or execute conditional response actions.
The greatest value of SOAR lies not in replacing analysts, but in reducing repetitive tasks, standardizing workflows, and providing sufficient context so analysts can make faster decisions. However, if workflows are flawed, input data is incomplete, or API permissions are overly broad, automation can propagate errors across multiple systems in a short time.
Therefore, organizations must classify operations by risk level: determining which steps can be highly automated, which should only be partially automated, and which actions require human approval. Playbooks must also incorporate validation checks, rate limits, logging, rollback capabilities, and emergency stop mechanisms.
What is SOAR in the SOC?
SOAR in the SOC is an orchestration, automation, and response layer that connects security tools with human-led workflows. SOAR stands for Security Orchestration, Automation, and Response, though these three components are not entirely synonymous.
Orchestration is the ability to integrate SIEM, EDR, XDR, identity systems, email, threat intelligence, ticketing, and other platforms into a unified workflow. The goal is to minimize the need for analysts to switch between multiple interfaces, copy and paste data, and repeat the same lookup sequence for every alert.
Automation is the system's execution of predefined steps, such as retrieving asset information, performing IP lookups, tagging incidents, or sending approval requests. Response refers to the remediation phase, which may range from simply providing recommendations to taking direct action, such as isolating endpoints, locking accounts, and blocking indicators.
This architecture must be positioned within the broader operational chain. A SOC architecture comprising logs, SIEM, SOAR, EDR, and XDR must clearly define which system handles detection, which manages orchestration, and who holds the authority to decide on response actions.
According to Microsoft Sentinel documentation on automation rules, these rules provide a centralized mechanism to manage automation - allowing teams to assign owners, change statuses, add tags, escalate incidents, or trigger playbooks.
According to the guide on creating and managing Microsoft Sentinel playbooks, a playbook is a collection of remediation procedures triggered in response to an incident, alert, or entity, and can be executed either automatically or manually.y tự động hoặc thủ công.
How does SOAR differ from SIEM, EDR, and XDR?
SOAR differs from SIEM, EDR, and XDR primarily in its role of orchestrating workflows, even though modern platforms often bundle multiple capabilities into a single product. Organizations should not evaluate these tools based solely on product category labels.
SIEM collects and analyzes logs, correlates events, and generates alerts or incidents. EDR focuses on telemetry, detection, and response at the endpoint level. XDR expands cross-layer signal correlation across endpoints, identity, email, cloud environments, and other layers depending on the platform.
SOAR ingests inputs from those systems, enriches the data, applies decision logic, creates tasks, and coordinates response actions. In other words, while a SIEM or XDR may detect an anomalous login, SOAR retrieves user context, checks asset details, searches for related events, creates a case, and sends an approval request to an analyst before locking the account.
In practice, these boundaries often overlap. A modern SIEM platform includes automation rules and playbooks; XDR can autonomously isolate endpoints; and SOAR tools feature built-in case management and threat intelligence. Organizations should evaluate integration capabilities, access controls, and workflow governance rather than purchasing additional tools simply for their category names.
Understanding the roles of EDR and XDR in the SOC helps define the detection and action layer on endpoints, while SOAR orchestrates this layer alongside other systems, workflows, and stakeholders.
Which SOC workflows are suitable for full automation?
Workflows with consistent inputs, clear logic, low operational impact, and verifiable outcomes are generally well-suited for high levels of automation. However, "full automation" still requires logging, error monitoring, and kill switches when anomalous conditions arise.
Suitable steps include case creation, data field normalization, deduplication, tagging, alert routing, asset enrichment, indicator lookups, telemetry collection, and sending notifications. These are repetitive tasks that require minimal human judgment, provided the data sources are reliable.

Certain incident management tasks can also be conditionally automated, such as assigning an owner, updating statuses, or adding analyst tasks. Microsoft Sentinel automation rules can be triggered when an incident or alert is created, updated, or has its properties modified, executing subsequent actions in a predefined order.
Auto-closing cases is only appropriate when the criteria are unambiguous, the impact is low, and the outcome is verifiable. For example, an alert that is an exact duplicate of an open incident can be linked or closed via rules. Conversely, a case should never be auto-closed simply because a single threat intelligence feed has not yet flagged an indicator as malicious.Ngay cả workflow ít rủi ro cũng cần owner. Even low-risk workflows require a designated owner. When an API changes, credentials expire, or a data feed stops updating, the system must generate an operational alert rather than silently skipping the enrichment step.
Which workflows should only be partially automated?
Workflows that require contextual evaluation, involve incomplete data, or branch into multiple possible outcomes should only be partially automated. SOAR can gather evidence and propose recommendations, while the analyst confirms the conclusion and determines the next steps.
Phishing email triage is a prime example. A playbook can extract headers, URLs, domains, and attachments, perform reputation lookups, run sandbox analysis, search for similar emails across the organization, and compile a list of recipients. However, an analyst must still evaluate the communication context, sender credibility, and potential evasion techniques.
MITRE ATT&CK classifies phishing as Initial Access technique T1566, encompassing sub-techniques such as malicious attachments, links, third-party services, and voice phishing. This breadth demonstrates that a playbook checking only URLs or attachments will fail to cover all attack variants.
Investigating anomalous logins also requires context. Automation can pull geolocation data, device details, user history, privilege levels, and post-login activity. Yet an analyst or the account owner must still verify business travel, VPN usage, or legitimate behavior before drawing a conclusion.
For malware incidents, a workflow can collect process trees, file hashes, network connections, and sandbox results. However, the decision to isolate a production server requires assessing business impact, failover capacity, and recovery plans.
The goal of partial automation is to bring the analyst to a decision point armed with actionable data, not to force the workflow all the way to an automated final action.
Which actions still require analyst approval?
Actions that are difficult to reverse, have broad operational impact, or involve critical assets still require approval from an analyst or an authorized stakeholder. Approval authority must be role-based, rather than relying merely on whether a user has permission to click a button in the tool.
Actions that typically require strict governance include isolating servers, locking privileged accounts, revoking tokens, purging emails in bulk, modifying firewall rules, blocking domains organization-wide, stopping services, or deleting data.
For standard user endpoints, organizations may allow automated responses if confidence is high and the action is reversible. However, executing the exact same action on a production server can cause severe business disruption. Consequently, playbooks must evaluate asset criticality before selecting a response branch.
Approval authority also does not always rest with the SOC analyst. Isolating an endpoint may require IT operations; locking a privileged account may require the IAM team; stopping an application may require the system owner; and external communications may require the CISO, legal, or PR teams.
NIST SP 800-61 Revision 3 places incident response within the broader context of enterprise risk management, linking Detect, Respond, and Recover activities to continuous improvement. This underscores that response decisions are not purely technical maneuvers - they must also account for recovery and organizational impact.
What components does a SOAR playbook need?
A robust SOAR playbook requires a trigger, input data, decision logic, actions, approval gates, verification steps, and a rollback plan. A playbook that only maps out a "happy path" without error handling is not ready for production.
A playbook should start with a clear trigger specifying the alert type, data source, and required fields. This is followed by the enrichment phase, which gathers asset details, user context, threat intelligence, and related events.
Decision logic must evaluate multiple factors rather than relying on a single indicator. These factors can include confidence scores, asset criticality, user privilege levels, scope of impact, and reversibility.
Every action needs a designated owner, timeout threshold, retry policy, and verification mechanism. If an API returns a success status while the endpoint remains connected to the network, the workflow should not treat containment as complete. If an action fails, the case must be routed to an analyst with full error details.
Microsoft allows Sentinel playbooks to run automatically or manually at the incident, alert, or entity level, depending on how the response workflow is designed. Manual execution serves as a practical control when organizations want to enforce an approval step prior to taking action.
How should a sample phishing playbook operate?
A sample phishing playbook can trigger when a user reports a suspicious email or when the email security gateway generates an alert. The workflow collects the message ID, sender, recipient, headers, URLs, attachments, and email authentication results.
SOAR then runs lookups on domains, URLs, and file hashes, searches for similar emails across the organization, checks whether users clicked any links, and builds a preliminary timeline. These findings are consolidated into a case for analyst review.
If the evidence is inconclusive, the analyst can request additional data or detonate the file in a sandbox. Once confirmed malicious, high-impact steps, such as bulk email purging, blocking indicators, or locking accounts, must pass through the appropriate approval gates.
Standard phishing investigation documentation can serve as the baseline for building a checklist; playbooks should only automate steps that the team has already validated and tested.
Where should Human-in-the-Loop be placed?
Human-in-the-loop checkpoints should be placed prior to decisions that require contextual judgment or carry the risk of significant operational impact. Inserting manual approvals at every step is unnecessary and defeats the purpose of automation.
There are four primary forms of human intervention: human review requires an analyst to validate enrichment results; human approval requires an authorized stakeholder to greenlight an action; human override allows personnel to halt or modify a workflow; and human escalation routes a case to senior specialists or management when conditions exceed defined thresholds.
Approval gates should be calibrated based on confidence levels, asset criticality, account type, scope of action, and reversibility. A standard user account and an administrative account should never follow the same automated lockout branch.
SOAR in the SOC also requires fallback handling when an approval request times out. The workflow can send reminders, escalate to a backup approver, or apply a lower-impact containment measure - it should never automatically escalate privileges or execute high-impact actions simply because a timer expired.
Approval authority must be clearly codified in the RACI matrix, tool-level roles, and runbooks. Simply receiving an alert notification does not grant an analyst the authority to accept business downtime.
What guardrails help prevent erroneous automated responses?
Effective guardrails combine access controls, scope limits, conditional checks, logging, and rollback mechanisms. No single guardrail is sufficient to protect an entire workflow on its own.
API permissions must adhere to the principle of least privilege. Service accounts used for enrichment lookups should not simultaneously hold permissions to delete data or isolate hosts. If a playbook requires high-risk actions, it should use separate credentials that are invoked only after approval.
Allowlists protect critical assets or accounts from automated actions. Rate limits cap the number of actions executed within a given timeframe. Circuit breakers automatically halt a workflow if error rates or the number of affected targets exceed predefined thresholds.
Idempotency ensures that re-running the same step does not produce unintended duplicate effects. For instance, a workflow retry must not generate multiple identical tickets or repeatedly send lock requests for an account that is already locked.
Dry runs log intended actions without executing them in production. Canary deployments roll out a playbook to a limited scope before wider adoption. A kill switch enables operations teams to immediately terminate a workflow if anomalous behavior is detected.
All playbook modifications require version control, peer approval, staging environments, scheduled review dates, and a designated owner. Any playbook lacking an owner or overdue for review should be flagged for audit rather than trusted indefinitely.
What operational risks does SOAR introduce?
The biggest risk of SOAR is automating a flawed assumption at scale. Errors can stem from faulty logic, incomplete data, overly broad permissions, broken integrations, or playbooks that no longer reflect the current IT environment.
Incomplete input data can lead to inaccurate severity scoring. A clean reputation score does not prove a domain is safe, and an outdated asset inventory can cause a mission-critical server to be treated like a standard user endpoint.
Over-privileged API credentials amplify the blast radius if a playbook is misconfigured or compromised. Expired secrets cause workflows to fail, while poor error handling can leave a case appearing resolved when containment actually failed.
Feedback loops can occur when a playbook's own action generates a new telemetry event that re-triggers the exact same workflow. Rate limits, unique correlation markers, and exclusion rules are essential to prevent infinite loops.
Vendor lock-in is another operational risk. Organizations must understand how to export playbooks, case histories, execution logs, and integration configurations if they migrate platforms. Inadequate documentation turns automation into a black box that is difficult to maintain.
Finally, misaligned KPIs can incentivize SOC teams to auto-close high volumes of cases rather than improving detection quality. The sheer volume of automated actions is meaningless if those actions do not demonstrably reduce risk.
To what extent can AI participate in SOAR?
AI can assist with incident summarization, content classification, investigative recommendations, and translating natural language into queries, but its outputs must be treated as suggestions requiring verification. AI should never be granted default authority to execute high-impact response actions.
While an AI-generated summary helps analysts review cases faster, it carries the risk of omitting critical details or hallucinating non-existent correlations. Every conclusion must trace back directly to underlying alerts, logs, telemetry, or raw evidence.
When AI processes emails, tickets, or external content, organizations must account for malicious inputs and prompt injection risks. Untrusted data must never be able to alter system instructions or trigger actions directly.
A secure architecture separates recommendation from execution: AI generates suggestions; deterministic logic evaluates conditions; an analyst or approval gate authorizes the action; and SOAR logs and verifies the outcome.
Organizations must also govern what data is sent to AI models, where it is processed, how long it is retained, and how it is used. Sensitive incident evidence should never be fed into an AI service without thoroughly vetting data privacy terms and security controls.
SOAR in the SOC can certainly leverage AI, but AI does not replace the human analyst accountable for context, evidence validation, and business impact.
Which KPIs accurately measure SOC automation effectiveness?
Effective KPIs must measure coverage, reliability, quality, speed, analyst value, and risk. Counting the number of playbooks built or cases auto-closed merely reflects tool usage - not operational effectiveness.
KPI Category | Example metrics | Governance question |
Coverage | Automated use cases, alerts with sufficient data, integration coverage | Is automation covering high-value workloads or only easy use cases? |
Reliability | Execution success rate, timeouts, API errors, retries, rollbacks | Do workflows run reliably when integrations change? |
Quality | Analyst overrides, rejected approvals, reopened cases, false containment | Is automation generating the right outcomes? |
Speed | Enrichment time, case creation time, approval wait time, escalation time | Which steps are saving time, and where are the bottlenecks? |
Analyst value | Cases with sufficient context, repetitive tasks eliminated, investigation time | Do analysts have more time for high-value analysis? |
Risk | Overdue playbooks, excessive permissions, out-of-scope actions | Is automation creating new risks? |
Every KPI requires a clear definition, an owner, a data source, and a baseline. A high execution success rate means little if the playbook only handles low-value tasks, and reduced handling time is counterproductive if it leads to a spike in reopened cases.
KPIs must be evaluated per use case. Phishing, anomalous logins, and malware involve different data sources, decision logic, and impact levels, so a single one-size-fits-all target should not be applied across all of them.
In particular, human approval wait time should be tracked separately. If automation completes enrichment rapidly but cases stall waiting for sign-off, the bottleneck lies in governance rather than technology.
Which use cases should organizations start with when deploying SOAR?
Organizations should start with use cases that occur frequently enough to matter, follow stable workflows, rely on clean data, and carry low-to-medium operational impact. Complex containment actions should never be chosen as the initial pilot just because they look impressive.
Indicator enrichment, case creation, tagging, owner assignment, and asset information gathering are typically ideal starting points. These steps are easy to measure and pose minimal risk of disruption.
Phishing is another common starting use case, provided the organization already has reliable email telemetry, sandbox tools, and threat intelligence feeds. The pilot should focus on ingestion, extraction, enrichment, and aggregation, while keeping bulk email deletion behind an approval gate until the workflow is proven.
Before building a playbook, the SOC team should observe the manual process, eliminate redundant steps, and map out exceptions. Automating an unstable process will only cause exceptions to occur faster and more frequently.
Organizations can leverage standard SOC incident investigation documentation to standardize inputs, evidence collection, and analytical steps before translating them into automated workflows.
Should organizations start with a pilot or a full-scale rollout?
Organizations should start with a pilot if they lack baselines for data quality, integration errors, and playbook behavior. A broad rollout is only appropriate once workflows have been thoroughly tested and have clear ownership, rollback procedures, and monitoring controls in place.
A pilot should be scoped to a single use case, a specific asset group, or a narrow set of actions. In the initial phase, playbooks can run in dry-run mode to log intended actions without executing them.
From there, organizations can expand automation in stages: starting with enrichment, followed by case management, notifications, approval-gated actions, and finally pre-authorized automated actions. Each expansion must be backed by empirical evidence of reliability and quality.
Once a playbook moves into production, the work is far from over. The SOC team must continuously review playbook versions, exceptions, access permissions, KPIs, and API changes. Security orchestration, automation, and response is an ongoing operational capability - not a one-time configuration project.
SOAR in the SOC only delivers sustainable value when it automates the right steps, keeps analysts embedded at critical decision points, and measures both speed and risk.
How can IPSIP Vietnam support SOAR assessment and deployment in the SOC?
IPSIP Vietnam can begin by assessing your current SOC workflows, alert sources, security stack, API permissions, and repetitive analyst tasks. These findings help prioritize use cases that deliver high value at an appropriate risk level.

The design phase defines triggers, inputs, decision logic, approval gates, response actions, verification steps, and rollback mechanisms. Both teams also establish ownership, access controls, testing environments, and KPIs.
Pilots should begin with dry runs or low-impact actions. Workflows are closely monitored for integration errors, missing data, analyst overrides, and approval wait times before expanding automated response permissions.
Specific integration capabilities and support scope must be validated against the platforms your organization currently operates. Playbooks should never be assumed to be portable out-of-the-box across different SIEM, EDR, XDR, or ticketing systems.
IPSIP Vietnam SOC 24/7
Explore our 24/7 SOC services to define your monitoring and incident response scope. If you are currently evaluating automation use cases, guardrails, or KPIs, schedule a consultation with IPSIP to review your existing tools, data, and workflows.

References
Create and use Microsoft Sentinel automation rules - Microsoft Learn
Create and manage Microsoft Sentinel playbooks - Microsoft Learn
Automate and run Microsoft Sentinel playbooks - Microsoft Learn
NIST SP 800-61 Revision 3 - National Institute of Standards and Technology, bản chính thức tháng 4/2025
MITRE ATT&CK Phishing T1566 - MITRE
Cortex XSOAR use cases - Palo Alto Networks











