Magrid by LetzMath Sarl
Why a number that large was available
Sixty five percent sounds like an exaggeration and I would treat it as one if I read it somewhere else. So here is why it was there to find.
Waste concentrates in environments where the person provisioning is not the person paying. Grant funded, investor funded, cross charged to another department, it does not matter which. The mechanism is the same. Nothing forces a second look, so the second look never happens, and two or three years of reasonable individual decisions compound into an environment that no longer resembles the workload running on it.
In a company paying its own way from day one, that same environment would have been questioned within a quarter. Here nobody had a reason to question it until the funding clock started running out.
If your infrastructure was stood up under a grant, a funding round, or another team's budget, the number sitting in yours is probably larger than you would guess.
Magrid by LetzMath Sarl
How we did it
One session with their CTO. We went through the resources one at a time and looked at what each one was actually doing against what it was provisioned for.
The App Service Plans were the big one. They were configured for a scale of load that was not arriving, and had been for a long time. Beyond that it was the usual accumulation. Virtual machines sized well above what they needed. SQL Server configuration that had never been revisited. Unattached disks left behind by resources that no longer existed, quietly billing by the hour. Service tiers nobody had reconsidered since the day they were first set.
None of it individually looked like a problem. That is the point. No single line on that bill would have made anyone stop. It was the total that mattered, and the total is not something you notice until you sit down and go through it deliberately.
We also put reserved instances in place for the workloads that were genuinely stable and going to keep running. That is a commitment, so it only made sense for the parts of the environment we had just established were the right size. Reserving capacity you have not right sized first is how you lock in the overspend for a year.
Their team applied the changes. Nothing broke.
Magrid by LetzMath Sarl
What this was, and what it was not
This was one working session with a CTO who knew his own environment well and could answer questions about what each resource was for. That is a considerable advantage and it is why a single session got this far. There was no written report and none was asked for.
I mention it because the audit I sell now is a different thing. Five days, a written ranked fix list, and a note on what breaks if a finding is actioned wrongly. That exists because most environments do not come with someone who can tell you in real time which VM is load bearing. When they do, the work goes faster.
Magrid by LetzMath Sarl
The part worth taking away
Cost work has a reputation for being about turning things off, and the risk in it is that you turn off the wrong thing. Going through the resources one at a time with someone who knows the system is slower than running a tool over it, and it is the reason nothing broke.
The other takeaway is the one about who feels the bill. If nobody in your organisation has a personal reason to care what Azure costs this month, the number is drifting, and the longer it drifts the more there is sitting in it.