TL;DR
IT change management practice was built for estates you could walk to. Stage it in a lab, pilot it on a floor, roll back overnight if it goes wrong. A distributed fleet breaks the middle and the end of that. There is no floor to pilot on, and rolling back a change on a laptop in somebody’s spare room requires that laptop to still be reachable, which is exactly what a bad change takes away.
- Rollback is the assumption that breaks. A bad change can remove your own access.
- Pilot on a deliberately chosen group, not on whoever updates first.
- Stage by risk to connectivity, not by department.
- Never ship a change that can break enrolment without a manual recovery path.
- Time zones mean there is no quiet window. Plan around that.
- Keep a known-good configuration people can be talked back into.
The assumption IT change management quietly makes
Every change process contains a rollback step, and it assumes the rollback will reach the device.
On a managed estate that holds. The machines are on a network you control, and if something goes badly wrong somebody can physically touch them.
On a distributed fleet the rollback path runs through the same management channel the change just modified. If the change breaks the agent, the VPN or the identity provider, the rollback has nowhere to travel.
The failure that actually happens
Not catastrophic breakage. The common version is subtler: a change that works fine on every machine that is online when it ships, and bricks the enrolment of the twenty that were off.
Those twenty come back a week later, fail to check in, and now require a human conversation to recover. Multiply that by a fleet across several countries and a routine change becomes a fortnight of support.
Stage by risk, not by department
The instinct is to roll out team by team, because that is how the organisation is shaped. It is the wrong axis.
The axis that matters is what happens if this change goes wrong on this device, which depends on connectivity, location and how recoverable the person is.
| Group | Who is in it | Why they go in this order |
|---|---|---|
| First | IT staff and volunteers | Can self-recover, will report clearly |
| Second | Office-based or near a hub | Physical recovery is possible |
| Third | Remote, well connected, same region | Recoverable by remote session |
| Fourth | Remote, distant time zones | Support overlap is limited |
| Last | Poor connectivity, travelling, field roles | Hardest to recover, least tolerant |
Most organisations do the opposite without meaning to. The rollout goes to whoever is online first, which skews toward the most connected and least representative users, and the problems surface at the end rather than the beginning.
Hold the last group longer than feels necessary
A week of soak time on the fourth group costs nothing and catches the failures that only appear on slower links or after a device has been offline for days.
Protect the channel you manage through
One rule prevents most serious incidents: no change ships that can break the mechanism used to ship changes.
That means the management agent, the identity provider integration, the network configuration and anything touching disk encryption get a different and more cautious process than application updates.
the NCSC device security guidance treats the ability to manage and update devices as foundational rather than optional, which is the right framing. A fleet you cannot reach is not a managed fleet regardless of what the console says.
Keep a manual recovery path
For the categories above, there needs to be a route back that does not depend on the thing that broke. A documented local admin procedure, a recovery key process, or in the worst case a replacement device.
That path should be tested occasionally rather than assumed. The moment to discover the recovery documentation is three years out of date is not during the incident.
the NIST Cybersecurity Framework is a reasonable structure for thinking about which changes carry this weight, though most teams need the single rule above more than they need a framework.
There is no overnight
Traditional change windows run at night because that is when nobody is working. A fleet spread across time zones has no such hour.
Three practical consequences.
- Changes must be interruption-tolerant. If it needs an uninterrupted hour, somebody will close the lid halfway through.
- Deferral has to be real. A prompt people can genuinely postpone gets accepted eventually. One they cannot gets worked around.
- Support has to be awake somewhere. Shipping a risky change into a region with no support overlap is a decision, not a schedule.
The second point is routinely underestimated. Forced immediate restarts during somebody’s working day produce exactly the behaviour you do not want, which is people avoiding the update channel entirely.
Our note on device health monitoring covers how to see the devices that are quietly falling behind.
The decision table
The weight of process should match the blast radius.
| Situation | Scale | Setup | Primary pain | Recommended starting point |
|---|---|---|---|---|
| Application update | Any | Managed deployment | Low | Staged rings, light process |
| OS major version | Any | Mixed hardware ages | Older devices fail first | Test by hardware generation, not team |
| Management agent change | Any | Single management channel | Can remove your own access | Slowest rings, manual recovery ready |
| Encryption or identity change | Any | Security-critical | Lockout risk | Pilot tiny, escrow keys, test recovery |
| Change during high churn | Any | Joiners and leavers | New devices out of step | Update the build image first |
Row five is forgotten constantly. If the standard image is not updated alongside the fleet, every machine issued during the rollout arrives in the old state and has to be caught up separately.
Verify afterwards, and count the stragglers
A change is not done when the console says ninety four per cent. It is done when the remaining six per cent has a name against each device.
Those stragglers fall into predictable buckets. Devices that are genuinely offline and will return. Devices with a person on leave. Devices nobody has used in months, which are usually with a former employee. And a small number that actually failed.
Separating those is what turns a stalled rollout into four short tasks instead of one vague one.
The ones with former employees
Every rollout surfaces devices that never check in again, and that list is one of the more reliable ways to find equipment that was never recovered. It is worth passing to whoever owns asset recovery rather than writing off as non-compliant.
For teams without local presence, those devices need collecting before they can be wiped or reissued. RemoAsset handles recovery from former employees and holds returned stock, which closes the loop a change rollout opened.
Our piece on asset register accuracy covers how to feed that back without creating a second parallel record.
Tell people what is coming
On a fleet you can reach, a silent change is fine because somebody can fix it. On a fleet you cannot, the user is your only diagnostic instrument, and an uninformed user reports symptoms rather than causes.
A short notice naming what is changing, roughly when, and what to do if something looks wrong converts a vague ticket into a useful one. It also stops people assuming the machine is broken and quietly working around it for a fortnight.
Keep it to three lines. Long change notices get ignored at exactly the same rate as no notice at all, and the goal is recognition rather than comprehension.
RemoAsset recovers devices that stop checking in, so a stalled rollout becomes a collection rather than a write-off. More at PeopleOpsHQ.
Frequently asked questions
What breaks about IT change management on a remote fleet?
Rollback. Standard change process assumes the rollback can reach the device, which holds on an estate you can walk to and fails when the recovery path runs through the same management channel the change just modified. If a change breaks the agent, the VPN or the identity integration, there is no route back. The common version is not catastrophic either, it is machines that were offline during the rollout failing to re-enrol afterwards.
How should a change be staged across a distributed fleet?
By recoverability rather than by department. Start with IT staff and volunteers who can self-recover and report clearly, then people near a hub where physical recovery is possible, then well-connected remote users in the same region, then distant time zones, and last anyone on poor connectivity or travelling. Most teams accidentally do the reverse, rolling out to whoever is online first, which is the least representative group.
Which changes need extra caution?
Anything that can break the mechanism used to deliver changes: the management agent, the identity provider integration, network configuration and disk encryption. Those deserve a different and slower process than application updates, plus a documented manual recovery path that does not depend on the thing that might break. Test that path occasionally, because discovering the recovery documentation is three years stale during an incident is the worst possible timing.
How do you handle change windows across time zones?
Accept that there is no overnight and design around it. Changes need to tolerate interruption, because somebody will close a lid halfway through. Deferral needs to be genuine, since a prompt people can actually postpone gets accepted eventually while one they cannot gets worked around. And shipping a risky change into a region with no support overlap is a decision rather than a scheduling detail.
What should happen to devices that never complete a rollout?
Separate them rather than treating them as one non-compliance number. They divide into devices genuinely offline that will return, devices belonging to someone on leave, devices nobody has used for months which are usually with a former employee, and a small number that actually failed. That third group is one of the more reliable ways to find unrecovered equipment, and it belongs with whoever owns asset recovery.
Should the standard build image be updated during a rollout?
Yes, and it is forgotten constantly. If the image used to provision new machines is not updated alongside the fleet, every device issued during the rollout arrives in the old state and needs catching up separately. In a company with steady hiring that quietly creates a second population running the previous configuration, which then shows as a stubborn compliance gap nobody can explain.
Is this the same as ITIL change management?
It borrows the idea and discards several of the assumptions. The ITIL framing grew up around estates and services where staging, piloting and rollback are all physically available. On a fleet of laptops in people’s homes the useful parts are the discipline of staged release and explicit risk assessment, while the change advisory structures matter far less than whether you can still reach a device after the change lands.
RemoAsset recovers devices that stop checking in, wherever the person who had them is.