A Windows audit policy baseline is only useful when the events it requests are actually generated, collected, parsed, and searchable. This guide provides a repeatable way to choose Advanced Audit Policy settings, estimate their operational cost, validate endpoint-to-SIEM coverage with benign tests, and document gaps before they affect an investigation.
Overview
Windows Advanced Audit Policy controls which security-relevant activities are written to the Security event log. A practical baseline should balance investigative value against event volume, storage, endpoint overhead, and analyst workload. Enabling every available category is rarely a good starting point; enabling too little leaves gaps in identity, process, policy, and access telemetry.
Configure Advanced Audit Policy through Computer Configuration > Policies > Windows Settings > Security Settings > Advanced Audit Policy Configuration > System Audit Policies. Apply settings through a deliberately scoped Group Policy Object (GPO), then confirm the resulting policy on representative clients and servers. Avoid relying on legacy “Audit Policy” settings and Advanced Audit Policy settings at the same time without testing precedence and resulting behavior.
The table below is a starting baseline. Event IDs are representative rather than an exhaustive reference; the exact fields and event behavior can vary by Windows role, version, policy scope, and action performed.
| Coverage area | GPO subcategory | Representative events | Benign validation action | Collection note |
|---|---|---|---|---|
| Logon activity | Audit Logon; Audit Logoff; Audit Other Logon/Logoff Events | 4624, 4625, 4634, 4647, 4648 | Sign in with a test account, sign out, and use an approved run-as test where appropriate | High value; failure events can become noisy during password or service-account issues |
| Account management | Audit User Account Management; Audit Security Group Management | 4720, 4722, 4726, 4738, 4728, 4729, 4732, 4733 | Create, modify, disable, and remove a disposable test account or group member in a lab OU | Usually moderate volume; particularly useful on domain controllers |
| Process execution | Audit Process Creation | 4688; 4689 where available and collected | Launch a harmless signed utility such as notepad.exe and record the initiating user | Potentially high volume; command-line collection and retention affect usefulness |
| Policy change | Audit Audit Policy Change; Audit Authentication Policy Change; Audit Other Policy Change Events | 4719, 4739, 4902, 4904, 4905 | Make and revert a documented lab policy change | Low volume, high investigative value; protect these logs from unapproved clearing |
| Object access | Audit File System; Audit Registry; Audit Removable Storage, as justified | 4656, 4663, 4660, 4658 | Access a deliberately monitored test file or registry key | Can be very noisy; use SACLs and narrow paths instead of broad collection |
| Privilege use | Audit Sensitive Privilege Use; Audit Non-Sensitive Privilege Use | 4672, 4673, 4674 | Perform a controlled administrative action with a test account | Prioritize sensitive privilege use and validate field quality before expanding |
| Directory service changes | Audit Directory Service Changes | 5136, 5137, 5139, 5141 | Change and revert a lab directory object when auditing is enabled on the relevant object | Most relevant on domain controllers; requires suitable directory auditing configuration |
How to estimate coverage and collection cost
Use a simple worksheet rather than treating the baseline as a binary enabled-or-disabled decision. For each subcategory, record four values:
- Coverage value: assign 0 if the telemetry is not needed, 1 if useful, and 2 if required for a defined detection or investigation workflow.
- Generation status: record whether the endpoint is configured for Success, Failure, both, or neither.
- Observed rate: count events from a representative endpoint during a fixed window, such as one business day. Use a lab or pilot population before extrapolating.
- Ingestion result: record whether the SIEM received the event, parsed its key fields, and made it searchable within the expected time.
A useful coverage percentage is:
Weighted coverage = completed required checks ÷ total required checks × 100
Count a check as complete only when the policy is present, a benign test produces the expected event, the collector forwards it, and the SIEM exposes the fields needed by the intended analytic. For cost planning, estimate daily event volume as observed events during the test window × target population ÷ tested population. Treat this as an estimate, not a promise: servers, workstations, domain controllers, and application hosts have different activity patterns.
Prioritize fields as well as event IDs. For process creation, review the new process name, parent process, account, logon ID, host, and command line if your approved logging configuration provides it. For logons, compare account, source address, authentication package, logon type, and workstation information. A received event that has lost its host or identity context may not support reliable detection.
Inputs and assumptions
Before deployment, document the assumptions behind the baseline:
- Asset scope: separate domain controllers, member servers, workstations, and sensitive application hosts.
- Policy scope: identify the GPO, OU, security filtering, inheritance, and any conflicting policies.
- Success and Failure choices: enable only the outcomes that support a defined use case, then test both where required.
- Retention and forwarding: record Security log size, overwrite behavior, collection agent, transport, index or table, and retention period.
- Normalization: map Windows fields to the SIEM schema and note vendor-specific names for host, user, process, source address, and event code.
- Test safety: use disposable accounts, lab objects, and approved utilities. Do not test credential dumping, destructive access, or unauthorized policy changes on production systems.
Keep endpoint output and SIEM output side by side. This distinguishes policy failure from collection failure and parser failure. A SIEM count of zero does not prove that Windows generated no events.
Worked examples
Example 1: process telemetry. A pilot includes 20 workstations. After a normal workday, five workstations are sampled and produce 8,000 event 4688 records in total. A rough daily estimate is 8,000 × 20 ÷ 5, or 32,000 events for the pilot population. Before enabling collection broadly, verify whether command-line fields are present, whether the SIEM parser preserves parent-child relationships, and whether the resulting volume is acceptable for the detection team. If only 4,000 of the 8,000 endpoint events arrive, investigate forwarding or filtering before tuning an analytic.
Example 2: account management. A lab test creates a disposable user, adds it to a test group, changes an attribute, and disables it. The endpoint or domain controller should produce the relevant account and group-management events. Compare the actor, target account, group name, domain, controller, and timestamps at each stage. If the Security log shows the events but the SIEM contains only an event code and message text, mark ingestion as present but analytic-ready as incomplete.
For each test, capture: test ID, host, policy version, action time in UTC, expected event IDs, observed endpoint events, observed SIEM events, field differences, ingestion delay, and disposition. This record becomes the evidence for a GPO change, SIEM parser update, or detection exception.
When to recalculate
Revisit the baseline after a new domain or OU rollout, Windows client or Server migration, GPO restructuring, endpoint-agent replacement, SIEM parser update, retention change, or material change in detection requirements. Also review it after an investigation exposes a missing identity, process, policy, or object-access field.
At least quarterly, rerun a small validation set across each asset class: successful and failed logon, process creation, account or group change in a lab scope, policy change, and monitored object access where enabled. Compare the results with the previous worksheet rather than relying on memory.
Use this deployment checklist:
- Define required use cases and asset classes.
- Configure a versioned, scoped GPO.
- Confirm the effective policy on pilot hosts with approved administrative tools.
- Run benign validation actions and record expected event IDs.
- Confirm forwarding, parsing, timestamps, identity, host, and process fields in the SIEM.
- Estimate volume and adjust retention, filtering, or SACL scope deliberately.
- Promote in stages and retain the before-and-after worksheet.
For the next step, compare this Windows baseline with your detection regression process in Detection Rule Regression Testing. For process-focused coverage, use the Sysmon Event ID Reference alongside native Security events; the two sources can complement one another but should not be assumed to contain identical fields.