Workstation management
Dividing settings between Group Policy and an endpoint management tool, the workstation standard (image, packages, BitLocker, user data) and Windows update deployment rings.
Group Policy and an endpoint management tool (for example Microsoft Intune, Microsoft Configuration Manager or another tool with an agent) can set the same values, and no console reports an error: the GPO is applied, the profile in the tool has the status “compliant”, and the workstation keeps the value written last. This document describes the division of settings between the two mechanisms, the standard for a workstation that can be rebuilt from an image and packages, and Windows update deployment rings as an example of an area where the choice of owner has practical consequences.
The ownership map
Every setting, application and user folder has a single source recorded in the ownership map together with a rationale that allows it to be verified years later. The “both can do it” rows matter most, because that is where conflicts arise.
| Setting | Owner | Why |
|---|---|---|
| Domain password and account lockout policies | GPO | apply at domain level |
| User rights, audit policies | GPO | must work from system start and depend on directory groups |
| Windows Firewall, domain profile | GPO | rules depend on the domain network |
| Folder redirection, drive mappings, printers by group | GPO | derive from the user’s identity |
| BitLocker and recovery key backup | both can do it, one is chosen | key in the directory or in the tool’s console, never two sources of settings |
| Windows update rings | both can do it, one is chosen | GPO is enough for workstations on the network, the tool when laptops work off it |
| Browser and office suite settings | both can do it, one is chosen | one owner for a whole branch of settings, not for a single value |
| Documents, desktop and pictures | both can do it, one is chosen | redirection to a share (GPO) or moving known folders to OneDrive, one mechanism per folder |
| Installing, updating and removing applications, third-party patches | tool | schedule, report, retry after failure; installation through GPO suits only the simplest cases |
| Inventory, compliance, remediation scripts | tool | a result and log from every machine; GPO has no central compliance report |
| Machines outside the corporate network | tool | Group Policy reaches them only with connectivity to a domain controller |
The boundary in the last row depends on connectivity: a laptop with an always-on VPN receives Group Policy, while a device joined only to Entra ID does not receive it at all. The tool has its price: the agent is another component to maintain, the licence is an additional cost, and changes apply on the tool’s schedule, not at sign-in. The decision for every “both can do it” row goes into the documentation with a date and the person who made it, because a map without dates no longer matches reality after a year.
How a conflict works
The mechanism is clearest on a single registry value. The setting Interactive logon: Machine inactivity limit is stored in HKLM\SOFTWARE\Microsoft\Windows\CurrentVersion\Policies\System as InactivityTimeoutSecs. If a GPO sets it to 900 seconds there, and a script from the tool’s agent running every few hours sets it to 600, the screen locks after fifteen minutes on some laptops, after ten on others, and on some one time or the other. Neither object is wrong; only the moment of the last write differs.
Group Policy applies at system start and sign-in, and then in the background every 90 minutes with a random offset of up to 30 minutes. A background refresh does not rewrite values every time, however: the client applies settings again only after detecting a change in its objects or in their list. The same happens after gpupdate /force, and security settings are by default additionally reapplied every 16 hours, even if nothing has changed.
| Event | When the value is written |
|---|---|
| System start, sign-in (GPO) | at every event |
| Background refresh (GPO) | every 90 minutes with an offset of up to 30 minutes, only after a change in the objects or their list is detected |
| Security settings (GPO) | additionally every 16 hours, even without changes |
gpupdate /force (GPO) | immediately |
| Endpoint management tool | on its own schedule, for example a script every few hours |
The state of a workstation therefore depends on which event occurred last, and the symptoms look random. Each side’s report truthfully describes what that side did, which is why no console signals a problem.
Tools that use the MDM channel store the state of their policies in HKLM\SOFTWARE\Microsoft\PolicyManager\current\device. For some settings there is a tie-breaking policy: ControlPolicyConflict/MDMWinsOverGP in the Policy CSP, available from Windows 10 version 1803. Once it is enabled, a setting delivered through the MDM channel wins over the same setting from Group Policy. The policy covers only settings from the Policy CSP, however, so it does not replace the ownership map, and the documentation advises against configuring the same settings in GPO and MDM outside its scope, because the outcome is then decided by a race. The state of a setting on a workstation is read from two places: on the Group Policy side from the gpresult report, on the MDM side from the PolicyManager branch, and each shows only its own write.
Changing the owner of a setting
Moving a setting from one mechanism to the other follows a fixed order:
- Source. In a
gpresult /hreport run as administrator, so that it includes the computer part, locate the setting and the object that delivers it (the Winning GPO column). In the Component Status section every extension should have the status Success. - Withdrawal from GPO. Remove the setting from the object. If it is a preference, first change the item’s action to Delete, wait until the workstations have applied it, and only then remove the item.
- Check on workstations. Confirm on a few workstations that the value has disappeared from the registry or returned to its default. Values from administrative templates in the
Policiesbranches disappear when the object stops applying, but Group Policy Preferences and some security settings, such asInactivityTimeoutSecs, stay in the registry permanently. The solution is then to enter the target value in the last GPO or to remove the value with the tool’s remediation script. - New owner. Add the setting in the endpoint management tool, first for a pilot group, and update the “what where” map.
The reverse order, adding to the new source first, guarantees a conflict during the transition, which often lasts longer than planned. Objects with preferences need extra patience, because step two ends only when all workstations have applied the Delete action, and workstations that are switched off or off the network do so last.
Housekeeping on the GPO side
Conflicts cannot be hunted in a domain where it is not known which Group Policy objects are active. Get-GPO -All and Get-GPOReport show when an object was changed, whether it is linked and whether parts of it are disabled; the status AllSettingsDisabled marks an object that does nothing. Every object has a description (the Comment field in the console, Description in PowerShell) with an owner and purpose, and the name says what it covers: WS - BitLocker (WS: workstations) or USR - Folder Redirection (USR: user settings). Before deletion an object is archived with Backup-GPO, which writes a copy to a given directory and changes nothing in the domain.
The review of objects is tied to the ownership map: an object whose scope has been taken over by the endpoint management tool is the first candidate for removal, and the archive allows it to be restored. A description with the owner shortens the review, because the question about an object’s purpose has an addressee and does not require reconstructing history from a report.
Workstation standard: image, packages and data off the disk
A workstation on which nothing exists solely on its disk is rebuilt from three sources: the image, policy-based configuration and application packages, while user data waits off the disk. Anything outside these sources is a manual change that vanishes at the first computer replacement and appears in no report. Each source has its own owner in the ownership map, so after a hardware failure nothing is established from memory.
Image. It contains the operating system, drivers for the models in the standard and the management tool’s agent, but not applications: an application in the image is updated together with it (for example once a quarter), and an application in a package whenever needed. Regularly refreshing the image spares a newly deployed computer from installing overdue operating system updates. An alternative is a manufacturer image with automatic registration in the tool based on a hardware identifier reported at purchase (for example Windows Autopilot): less work on the image, less control over what the manufacturer installed. When configuration is to come from Group Policy and BitLocker keys are to go to AD DS, Autopilot has to run in the Microsoft Entra hybrid join variant, which Microsoft does not recommend for new devices; a device joined only to Entra ID does not receive Group Policy.
Hardware and names. Hardware is limited to two or three models, because every additional one is a separate set of drivers and a list of known issues. The computer name follows a pattern (PC-0142 for a desktop, NB-0217 for a laptop), and the number leads to a record in the inventory: model, serial number, user, deployment date.
Application packages. They are assigned to directory groups, not to users: GG-APP-Base for everyone, GG-APP-Accounting, GG-APP-Designers. A new employee of a department receives the applications when added to the group, and a move to another department means a group change and uninstallation of packages. An application outside a package is an exception with an owner and a review date, and the difference between the standard and reality is shown by the inventory from the tool’s console or one built from the Uninstall registry keys.
User data. Documents, desktop and pictures are redirected to a network share (Group Policy, Folder Redirection) or synchronised with the cloud, for example by moving known folders to OneDrive. These are two mechanisms for the same folders, so one mechanism per folder applies, recorded in the ownership map; configuring both at once means a conflict like the one with InactivityTimeoutSecs. Redirection to a share requires access to the corporate network, so on laptops outside the office it needs a VPN or offline files, while cloud synchronisation works over the internet. A path through a DFS namespace (\\corp.example.com\shares\users) instead of a server name allows data to be moved without changing the policies. After deployment nothing remains on the disk whose loss would be painful, and on computer replacement the data appears after the first sign-in.
BitLocker: key in the directory before encryption
Disk encryption illustrates the single-source principle well, because the recovery key has to be stored outside the computer before the computer fails. The Group Policy path (names from the English console):
Computer Configuration
└ Policies
└ Administrative Templates
└ Windows Components
└ BitLocker Drive Encryption
└ Operating System Drives
└ Choose how BitLocker-protected operating system drives can be recovered
[x] Save BitLocker recovery information to AD DS for operating system drives
[x] Do not enable BitLocker until recovery information is stored to AD DS for operating system drives
The last option is the most important. Without it, encryption can be enabled on a computer that has not stored the key in the directory, and after a motherboard failure with a TPM module such a disk is unreadable. The policy stores the key when encryption is enabled, so on disks encrypted earlier the presence of the key in the directory (msFVE-RecoveryInformation objects under the computer object) requires a separate check. A missing key can be backed up with manage-bde -protectors -adbackup C: -id <identifier>; protector identifiers are shown by manage-bde -protectors -get C:. The policy-based configuration also includes a unique local administrator password (Windows LAPS) and updates in rings.
Windows update deployment rings
A cumulative update installed on all machines at once breaks something on all machines at once, while not updating at all leaves hundreds of machines with vulnerabilities known for months. Rings spread the risk across stages, each of which can be stopped.
Four rings
| Ring | Who | Share | Quality update deferral | Enforcement deadline | Grace period |
|---|---|---|---|---|---|
| 0 | IT team | about 5% | 0 days | 2 days | 1 day |
| 1 | pilot: volunteers from every department | about 10% | 4 days | 3 days | 2 days |
| 2 | everyone else | about 80% | 10 days | 4 days | 2 days |
| 3 | sensitive: production, certified configurations | about 5% | 14 days plus a pause until approval | after the owner’s approval | 2 days |
The deferral counts from the update’s release: a device is offered the update only after the given number of days has passed (the limit for quality updates is 30 days). In Windows 11 the enforcement deadline counts from the moment the device discovered the update in a scan, and the grace period from the moment the device is waiting for a restart after installation; the effective deadline is the later of the two. A workstation that installs the update right after discovering it is therefore restarted at the latest 2 days after release in ring 0, 7 in the pilot and 14 in ring 2. A workstation that installs only at the deadline gets the grace period on top, which gives 3, 9 and 16 days respectively.
A pilot needs coverage: every critical application and every hardware model has at least one workstation in ring 1. Ring 3 covers workstations controlling production and machines with a certified software configuration; updates reach them after the owner’s approval, in an agreed window. For such devices Windows 11 version 24H2 and later also has maintenance windows, available since January 2026, which restrict downloading, installation and restart to set hours; the documentation recommends configuring either maintenance windows or enforcement deadlines, not both together.
Rings in Group Policy
Each ring is a separate Group Policy Object with the same structure and different values, filtered by a computer group: GG-WU-Ring0 to GG-WU-Ring3. The settings (names from the English console and the Windows 11 templates):
Computer Configuration
└ Policies
└ Administrative Templates
└ Windows Components
└ Windows Update
├ Manage updates offered from Windows Update
│ ├ Select when Quality Updates are received → deferral in days, pause
│ ├ Select the target Feature Update version → target operating system version
│ └ Do not include drivers with Windows Updates → drivers through a separate channel
└ Manage end user experience
├ Configure Automatic Updates → automatic download and install
├ Specify deadline for automatic updates and restarts for quality update
└ Turn off auto-restart for updates during active hours → active hours, e.g. 7–19
In Windows 11 version 22H2 and later the deadline policy (enforcement deadline and grace period) exists separately for quality updates and for feature updates; the single Specify deadlines for automatic updates and restarts policy applies to Windows 10. The target version setting keeps workstations on a proven operating system version until the pilot has tested the next one. The version has to be updated before end of servicing, because a workstation is updated automatically once it is 60 days past that date.
A gate, not a calendar
Moving to the next ring is a decision, not the passage of time.
The condition is an absence of update-related reports from rings 0 and 1 during the deferral period of ring 2. A report means the rollout is paused for the following rings until diagnosis. For this the setting Select when Quality Updates are received has a pause start date field (Pause Quality Updates starting): entered in the ring 2 object, it stops quality updates for the whole group for up to 35 days from the given date.
The enforcement deadline is a separate parameter and it determines effectiveness: without a deadline the share of updated machines usually stalls at a level where some users postpone the restart indefinitely. Until the deadline the restart takes place outside active hours (a range of at most 18 hours; the documentation advises not setting them by policy if users can set their own), and after it passes the restart is forced regardless of them.
Laptops outside the office
Updates distributed only from a server on the corporate network never reach a laptop that connects to the VPN for a quarter of an hour once a week. After a year it is one of the most vulnerable machines in the organisation, and at the same time one that often works on public networks. Windows Update for Business policies do not require a server on the network: the workstation downloads updates directly from Microsoft over the internet, and in the office Delivery Optimization limits traffic on the link. Policies already written to the registry keep working, also without a domain controller, but a laptop that has not seen one for months does not receive changes: a new deadline or a pause. For such machines the rings are run in the endpoint management tool, in line with the ownership map.
Who has fallen behind
The state of a workstation is assessed by the cumulative update number (UBR) from the registry (HKLM:\SOFTWARE\Microsoft\Windows NT\CurrentVersion, values CurrentBuild and UBR): after the ring 2 deadline workstations should have the latest UBR of the current month, and an older number marks machines that have fallen behind. The ProductName value is unsuitable for recognising the system, because in Windows 11 it still contains “Windows 10”; the version is recognised by CurrentBuild (22000 and higher is Windows 11). A pending restart combined with a last boot more than a dozen days ago means that the enforcement deadline is not working or the machine never restarts, only sleeps.
A separate list consists of workstations that do not check in to the domain: machines withdrawn from use or working permanently off the network, whose state must come from the last-contact report in the management tool. A device that does not check in is a bigger problem than one with an installation error, because an error at least is known.
Decisions and alternatives
| Approach | When it makes sense | Limitation |
|---|---|---|
| Group Policy as the owner of settings | workstations permanently on the domain network or with a working VPN; settings that depend on identity and groups | no central compliance report; no changes received off the network |
| Endpoint management tool as the owner | laptops working off the network, applications, inventory, remediation scripts | agent to maintain, licence cost, application on the tool’s schedule |
MDMWinsOverGP as conflict resolution | settings from the Policy CSP delivered through the MDM channel | covers only settings from the Policy CSP; does not replace the ownership map |
| Own image with packages | full control over workstation contents | image maintenance (refreshed for example once a quarter) |
| Manufacturer image and Windows Autopilot | less work on the image, registration based on a hardware identifier | less control over contents; Group Policy and BitLocker keys in AD DS require hybrid join, not recommended for new devices |
| Folder redirection to a share | workstations on the corporate network, data in one place | needs a VPN or offline files outside the office |
| Known folders in OneDrive | laptops outside the office | synchronisation over the internet; one mechanism per folder |
| Rings in GPO | workstations permanently on the domain network | laptops that do not see a domain controller do not receive changes |
| Rings in the management tool | laptops outside the office | dependence on the agent and its last-contact report |
Implementation
- Inventory of Group Policy objects (
Get-GPO -All, status, modification date, statusAllSettingsDisabled) and agpresult /hreport from several workstations: the winning object for every setting and the status Success for all extensions. The result feeds the “what where” map, and objects without an owner receive a description or are archived withBackup-GPO. - Recording the decisions for the “both can do it” rows with a date and a person. Moving settings between mechanisms in the order: source, withdrawal from GPO, check on workstations, new source, always first for a pilot group.
- Defining the workstation standard: two or three hardware models, a naming pattern, an image with the tool’s agent, application packages assigned to
GG-APP-*groups, user data off the disk (one mechanism per folder). The application inventory from the tool’s console or from the Uninstall keys shows the difference from the standard. - Enabling the BitLocker policy with the option that delays encryption until the key is stored in AD DS. Check: the disk encrypted with protection on (
Get-BitLockerVolume) and anmsFVE-RecoveryInformationobject under the computer object in the directory. - Creating the groups
GG-WU-Ring0toGG-WU-Ring3and four Group Policy objects with the values from the ring table. Ring 1 covers every critical application and every hardware model. Check: deferral, deadline and grace period values in theHKLM:\SOFTWARE\Policies\Microsoft\Windows\WindowsUpdatekeys match the row of the ring table. - Finding WSUS leftovers (the policy Specify intranet Microsoft update service location) and moving the rings of laptops outside the office to the endpoint management tool.
- Operating on a monthly cycle: moving to the next ring after an absence of reports, comparing UBR after the ring 2 deadline, reviewing the list of devices that do not check in.
Limitations and pitfalls
Win32_Productin inventory. The query runs a consistency check of every installed MSI package, takes minutes, loads the disk and can reconfigure installed applications. The Uninstall keys serve for the inventory: fast and read-only.- Workstation without a domain controller. A computer that cannot see a domain controller keeps the policies it last downloaded: a laptop working remotely without a VPN can have configuration from months ago, and no console shows an error. The last policy application date in the
gpresult /routput reveals such a case. - Computer in two ring groups. It receives both objects, and without a warning the winner follows processing order: in the same container the object with the lower link order (higher precedence), across containers the position in the hierarchy and the Enforced option. A change to a computer’s group membership takes effect only after a restart, because group membership goes into the Kerberos ticket at system start. After removing Authenticated Users from security filtering the computer needs read access to the object, so the ring group must have Read only on the Delegation tab.
- WSUS leftover. If Specify intranet Microsoft update service location remains in any object, a Windows 11 workstation downloads all updates from the WSUS server until the scan source policy (Specify source service for specific classes of Windows Updates) points to Windows Update. Deferrals from the rings do not change the source, and the former dual scan policy is not supported in Windows 11.