A thank you to FinOpsly, an early supporter and paid partner of FinOps & Beyond. They cover AI economics as well as cloud spend. I may bring them to my current client when the time is right, but we're working on more basic controls first.

Last week I wrote about getting tagging right, which really means getting the data right. Then the client's IT Director asked for projected annualized savings, something they could share with the CFO and CIO. I shared potential numbers, and they liked what they saw. His next question was when they would start to see the changes.

So we're running the tagging work and the savings work in parallel. That takes more effort. Before we change anything and because the current tags are ambiguous, we have to learn who owns each resource and what a change will touch.

We're applying tag names/keys with Azure Policy at the subscription level, tag values on existing resources at the Resource Group level, and then when all is done, we will apply deny and lookup rules at the subscription level. The policy work itself isn't hard. Finding the owners and ensuring we have the right current values is the time consuming part.

For ownership, every resource group and subscription needs an owner contact, ideally a distribution list, so a cost question or budget alert has somewhere to go. Some teams don't have a list. Some have one and are reluctant to share it. Some have a list nobody reads and we are attempting to avoid the DL blackhole. We started with none, even though some subscriptions were clearly marked for certain teams, but that doesn’t always help.

We're also being careful. The change has to be non-impacting, so every policy gets tested before it goes wide. And we are only adding tags. Existing tags stay as they are. Cleaning those up comes later, once the new standard has proven itself.

That's why it's slow. (And way slower than I want.)

In parallel, we're going after the easy savings. Orphaned disks. Stopped VMs. Disks attached to VMs that are deallocated or stopped. Stale snapshots. We expect all of this to save in the high, 10’s of thousands, to the low, 100’s of thousands in annualized savings.

None of this is clever. Every estate has it. It's the first thing a tool or a consultant (like me) finds, and it makes the first savings number easy to put in front of an executive.

Here's the catch. Delete a disk and you need to be sure nobody wants it. How do you know? You ask the owner. And if the resource has no owner tag and no contact, you either guess, or you wait. I have done both in my career, and waiting is cheaper than guessing.

Then there's the part nobody puts on the slide. Orphaned disks come back. A deleted snapshot gets recreated by the same job that made it. A stopped VM gets stopped again next quarter. A cleanup is a one-time event on a system that keeps producing the waste. That’s why its critical we know the owners and implement a system that will keep catching and removing the waste.

Savings and ownership are “2 peas in the same project”. The tag helps to tell you who owns the thing and the distribution list tells you who to ask. The policy makes sure the next resource shows up with both. Take any one away and the quick win turns into a quarterly cleanup that somebody has to remember to run.

That's governance. Not a committee. This is why every resource has an owner, making sure there's a standard for what "owned" means, and automation that enforces the standard without anyone remembering to.

I believe the order of operation is data, ownership, savings. We're doing the last one early because we have to. It works because the first two are underway.

This client isn't engaging with AI spend yet, and none of its software products have AI built in. When AI spend does show up, none of the resources in an AI estate will arrive with an owner unless we build the habit now. Most of the tags aren't AI-specific (that was the point of Issue 23). It will arrive through procurement, through an API key in a product feature, and through someone's personal plan on an expense report (Issue 30).

If we can't find a distribution list for a resource that has been running for years, we won't find an owner for a model call made next year. Build the owner field and the policy now, on the cloud estate you already have, and AI spend has somewhere to land when it arrives.

Pick the one owner field you can't live without and make it mandatory. A distribution list, not a person. People leave. Lists survive.

Run your quick wins through the owner field before you delete anything. No owner means a ticket to find one, not a guess. Record how many resources had no owner. That number is your governance gap, and it belongs in the savings report next to the savings.

Policy, policy, policy. Use what the cloud providers give you. Otherwise, build automation. Don’t make this move a calendar reminder for quarterly cleanup.

Remember - cleanup finds the savings. Governance keeps them.

Zip published its Enterprise AI Index last month, written by Nick Heinzmann. It looks at about $18 billion in approved purchase requests over 48 months from their customers. A few numbers I found interesting:

My read. The high rejection rate is companies using the one governance gate they have, procurement, and using it harder. The dataset can't say why requests failed (Zip says that too), so I won't guess at the reasons. But the concentration is useful. If one vendor holds 68% of your AI spend, the first governance job is small: one owner, one budget, one set of tags, on that vendor. You don't need to govern all six to cover most of the money.

In any case - thanks for reading this far. Have a great week!

Written with the help of AI. All the ideas expressed are mine and mine alone.

Company: FinOpsly Website; FinOpsly Directory Listing

Category: Software

What They Do: FinOpsly is an AI-native Value-Control platform helping enterprises map cloud, data, and AI spend to business value and automate cost control through agentic AI

Why It Matters: Cost governance is becoming critical in the AI era