Azure Virtual Desktop Consultancy

What does an AVD consultancy engagement actually involve?

Three phases, run as separate, honestly scoped pieces of work rather than one undifferentiated project. The assessment comes first and is deliberately cheap to say yes to: we look at what you have or what you're planning to build, and hand back a written recommendation covering platform fit, host pool design, storage, image strategy and a genuine cost projection, not a sales deck. Nothing gets built on the strength of a hunch.

When is AVD more cost-effective than Windows 365, and when isn't it?

AVD wins on cost at scale and with variable demand, because pooled multi-session hosts and autoscale mean you stop paying for a seat the moment nobody's using it. A host pool that ramps down overnight and at weekends can genuinely halve its own footprint compared with the hours it covers at peak, and that saving compounds across every user sharing the pool.

It also wins outright wherever the workload needs something Windows 365 doesn't offer at all: GPU-backed hosts for CAD or rendering, or fine-grained control over VM sizing that a fixed Cloud PC tier can't match.

Windows 365 wins where headcount is stable and nobody wants infrastructure decisions: a fixed per-user, per-month price with no pool to size, no scaling plan to tune, and a bill that doesn't move with Azure's own pricing changes. The honest crossover point isn't a fixed number of users, it's whether your concurrency is predictable.

A team that's logged in nine-to-five, every seat, every day, gets little benefit from AVD's ability to scale down, so the design overhead buys you less than the simplicity Windows 365 offers instead. We size this against your real login data at assessment rather than a rule of thumb, because we've seen the crossover point sit in very different places for two businesses of identical headcount.

What does a genuine AVD cost optimisation checklist cover?

In the order we work through it: autoscale first, because a scaling plan that ramps host pools down out of hours and at weekends is the single biggest lever and it's a configuration change, not new infrastructure.

Then VM SKU right-sizing, because pools are routinely left at whatever size the original deployment guide suggested and run under half capacity at peak. Then session density, checking whether users who'd happily share a multi-session host have instead been given dedicated ones out of caution rather than need.

Licensing is the final check: AVD access rights bundled into Microsoft 365 E3/E5, Business Premium or a standalone per-user licence are easy to duplicate-purchase without anyone noticing.

How do you size host pools and scaling plans without guessing?

From measured peak concurrent sessions, not headcount. Almost no estate has every user signed in at once, so sizing a pool to total staff bakes in permanent overspend from day one. We look at the actual shape of the working day: when sessions start, when they peak, how long they run, and what proportion of users are ever concurrent, and size the pool to that pattern with headroom rather than to a worst case that rarely happens.

VM SKU and user density then come from the application set that pool actually runs, because a browser-and-Office user and someone running a memory-hungry line-of-business client aren't the same workload and often shouldn't share a host at all.

The scaling plan is built around the same data: ramp-up ahead of the real login wave, a peak period sized to actual concurrency rather than headcount, and ramp-down that's aggressive enough to matter but doesn't force users to wait for a host to spin up.

What's genuinely different about managing AVD versus managing Windows 365?

Windows 365 management is close to managing a subscription: assign or remove a Cloud PC licence, and Microsoft handles the underlying host. AVD management is closer to managing infrastructure you happen not to own physically: host pools, scaling plans, image versions, storage performance and network configuration are all decisions you make and keep making, because none of them stay correct on their own as usage changes.

That's not a criticism of AVD, it's the trade for the control and cost efficiency it offers at scale. What it does mean is that AVD management needs an owner, whether that's an internal team with the time to do it properly or a consultancy running it as an ongoing service.

An AVD estate left to run itself after go-live is the single most common reason costs and logon times both drift, because nobody revisits the decisions that were correct when they were made and stopped being correct six months later.

How does multi-session change the blast radius of an incident?

It widens it, and that belongs in the decision alongside cost. A pooled multi-session host runs several users on one operating system instance, which is exactly where AVD's density and cost advantage comes from. It also means an incident on that host is an incident for everyone signed in to it, rather than for one person.

Concretely: a malware detection sits on the same OS instance as its neighbours' sessions. One user's runaway process degrades performance for everyone sharing the host. A regression introduced by an image update reaches every session landing on the affected hosts.

Taking a host out of service means draining it and moving live users first. By contrast, the equivalent event on a Windows 365 Cloud PC, or on a personal AVD host pool, affects one user and is contained by reprovisioning one machine.

Where a workload is regulated, or where per-user isolation is easier to evidence to an auditor than a set of compensating controls, personal host pools or Windows 365 Cloud PCs are often the better answer, and we will say so rather than defaulting to whichever platform the engagement started with.

FSLogix and storage: what actually goes wrong?

Almost always one of two faults. The first is storage sized for capacity rather than concurrency: FSLogix mounts a per-user profile container at sign-in, so a share that copes fine with a trickle of logins collapses under a simultaneous morning wave, and the fix isn't more storage, it's storage tiered and split for the concurrency you actually see.

The second is containers that grow without limit because nobody set exclusions, so Teams and browser caches, OneDrive cached files and search index data end up roaming inside the profile, slowing every mount and steadily filling the share.

We test against a real login wave rather than a single afternoon test session, because one login at 2pm proves nothing about what happens at 8:55 on a Monday. Housekeeping for containers left locked after an ungraceful disconnect is part of the same problem and is usually the difference between profiles that stay fast for months and ones that quietly degrade.

How should applications reach an AVD session host?

Three routes, and a well-run estate uses all of them deliberately rather than defaulting to one. Baked into the golden image, for agents, baseline configuration and anything that genuinely must be present at boot. Installed per host through Intune, for applications that don't need to be in every image build.

Attached at sign-in with MSIX App Attach, which expands a package from an SMB share and registers it per user without ever installing it on the host, keeping the image thin and letting different teams get different applications from the same pool.

The skill is in the routing decision, made per application rather than as a blanket policy. Anything that updates often should never live in the image, or every release becomes a rebuild and redeploy across every host pool that uses it. MSIX packaging and App Attach image builds are frequently the same underlying work as our dedicated application packaging service, and estates running both usually save by scoping them together.

Can you take over an AVD environment someone else built?

Yes, and it's one of the most common reasons we're brought in. The trigger is usually the bill, the logon times, or the person who originally designed the environment having moved on with nobody left who knows why it's configured the way it is. We read what's there before changing anything: host pool and scaling plan configuration, image lineage, FSLogix and storage design, licensing, Conditional Access and application delivery.

Plenty of what we find is sound, and we say so rather than rebuilding for the sake of it. What's usually wrong is a handful of early decisions rather than the whole build: a scaling plan that was never turned on, an image that's grown unmanaged for two years, storage that was sized once and never revisited. Fixing those is a fraction of the cost of a rebuild and usually delivers most of the saving.

What should you expect from an Azure Virtual Desktop provider?

Three things, and it is worth asking for all of them in writing before anyone quotes. First, a sizing method: ask how the provider arrives at host pool size and VM SKU, and be wary of any answer that starts from headcount rather than measured peak concurrency.

Second, who owns the environment after go-live, because AVD is a platform that keeps needing decisions and most virtual desktop providers price the build and leave the tuning unowned. Third, whether they will tell you not to buy it. A provider whose Azure Virtual Desktop services only ever conclude that you need Azure Virtual Desktop is describing a product line, not giving you advice.

The Microsoft partner badge on its own is thin evidence here. Almost every reseller carries one, and hosted virtual desktop providers vary enormously in how much AVD-specific engineering sits behind the logo.

The questions that separate them are practical: can they show you an FSLogix storage design they have argued against on cost grounds, do they run scaling plans they built a year ago and still review, and will they hand back an assessment that recommends Windows 365 for part of the estate when that is the right answer.

Does Azure Virtual Desktop make sense for a small business?

Often not, and that is worth saying plainly on a page that sells it. Azure Virtual Desktop for small business tends to disappoint below roughly twenty to thirty desktops, because the design and management overhead is close to fixed no matter how many users share it, while the pooled-host saving that pays for that overhead scales with user count. Under about twenty users the sums usually favour Windows 365, where nothing has to be sized and the price is known in advance.

The exceptions are real though, and they are about capability rather than headcount. A ten-person design studio needing GPU-backed hosts has no Windows 365 option at all. A small team with wildly variable seasonal demand can still benefit from autoscale. And a business with one specialist application that needs a particular VM series will find AVD is the only platform that offers it. We will tell you which case you are in at assessment rather than after the build.

Is Azure Virtual Desktop included in Microsoft 365 Business Premium?

The Windows licence is, the Azure bill is not, and conflating the two is the most common budgeting mistake we see. Microsoft 365 Business Premium is an eligible licence for Azure Virtual Desktop, alongside Microsoft 365 E3, E5, F3, A3 and A5 and the standalone Windows Enterprise E3 and E5 SKUs. If your users are already on Business Premium, they already carry the AVD access rights and you should not be buying them again.

What Business Premium does not cover is the infrastructure. Every session host is an Azure virtual machine you pay for by the hour it runs, plus the storage behind FSLogix profiles, plus egress and any networking. That is the part that moves, and the part a licence-led conversation tends to skip.

If you are running Windows Server session hosts rather than Windows multi-session, RDS CALs come into it as well. We check all of this against your existing tenant at assessment, because duplicate-purchased AVD rights are one of the easier savings to find.

Which AVD management and automation tools are worth it?

Native Azure tooling covers more than it used to. Scaling plans, the Compute Gallery and Azure Monitor between them handle autoscale, image versioning and alerting without a third-party product, and for a single host pool with settled working patterns that is often the whole answer. The case for a management layer starts when you are running several host pools, rebuilding images regularly, or need cost reporting that finance will actually accept.

Three are worth knowing about, and they solve different problems. Nerdio is the best known and is squarely an AVD and Windows 365 management layer, wrapping autoscale, image lifecycle and cost controls in one console. Hydra by Login VSI comes at it from the testing and user-experience side, which matters most when you need to prove an image or an application change is safe before it reaches the estate.

EtherInsights by EfficientEther works on the analysis and cost side, and it is the tooling behind the TCO modelling we use at assessment. EfficientEther is a sister company rather than an arm's-length vendor, and that is where the tooling came from: the same cost and capacity analysis, done by hand often enough that it was worth building properly.

Our position is that tooling is a multiplier on a good design and no substitute for one. A management layer applied to host pools sized from headcount rather than measured concurrency automates the wrong configuration faster. We size and tune the estate first, then recommend tooling if the ongoing operational load justifies the licence, and we will say when it does not.

The credential behind the recommendation

Ryan Mangan, Systech's founder, is a Microsoft MVP, an award renewed specifically for his work in Azure Virtual Desktop and Windows 365, and the author of Mastering Azure Virtual Desktop, published by Packt across two editions.

He maintains a public MSIX App Attach reference wiki that other engineers use when they get stuck on exactly the failure modes described on this page. The wider team is certified to AZ-140, Microsoft's Azure Virtual Desktop specialty exam, so the engineers who build and run your environment hold the platform credential rather than a general cloud one.

What that's worth to you is narrow but real. Reference material gets written by people who have already hit the failure modes and had to explain them clearly enough for someone else to fix theirs. The recommendations here aren't a generalist's best guess at a platform they cover alongside a dozen others, they come from the person who documented how AVD goes wrong for a living.

If you are comparing Microsoft AVD partners, that is the specific thing worth comparing: not who holds a badge, but who has had to explain the platform to somebody else in public and be right about it.