What Azure Advisor does not do

Advisor is better than people think

Microsoft’s own reference list of Advisor cost recommendations is long. It covers underutilized virtual machines and scale sets, unattached managed disks, snapshot storage tiers, reservation and savings plan opportunities across a dozen services, App Service plans with no apps running, Cosmos DB autoscale, and several AKS recommendations including vertical pod autoscaler mode and spot node pools. If someone tells you Advisor does not find waste, they have not read the list.

The gap is not detection. It is everything after.

Advisor produces a recommendation and stops. Nobody is assigned. Nothing is approved. No one records whether the change actually happened, and no one checks the next invoice to confirm the saving arrived. In most mid-market tenants the recommendations pile up until the list is long enough to feel hopeless, and then it gets ignored.

That is not a tooling problem you fix by buying another dashboard. It is an ownership problem. A finding becomes a saving only when a named person accepts it, someone else approves it, the change runs somewhere safe, and a later scan confirms it.

Three things the cost list does not include

Read the cost recommendation reference end to end and three ordinary sources of mid-market waste are absent. Azure Hybrid Benefit that was never applied to eligible Windows Server and SQL Server workloads. Development and test workloads running 24 hours a day because nobody set a schedule. SQL elastic pools and geo-redundant storage tiers sized for a peak that never came back.

None of these are exotic. They are the ordinary result of a tenant that grew faster than anyone had time to review it.

The numbers are estimates, and Microsoft says so

Advisor is explicit that its figures are potential savings. Reservation recommendations state that savings are estimated per subscription. Several right-sizing recommendations are based on the past seven days of usage, which is a short window for anything seasonal. That is a reasonable starting point. It is not a result you can hand to your finance team.

The number that survives a finance conversation is the one netted against the invoice you actually received. That is a different number, and getting to it is the work.

What to do with Advisor

Keep it. It is free, it is already running, and it is a reasonable first sweep. Treat its output as raw input to a process rather than as the process. Then add the four things it does not do: an owner on every finding, an approval before anything changes, a safe place to run the change, and a verification against the real bill.

That is the loop. Advisor tells you what. Someone still has to make it happen, and prove it.