This post is for the person who owns the budget line on a classified environment - the program manager, or the IT lead who runs it - and who has just been handed a specific instruction.
Someone above you has looked at the utilization figures. They have seen that the environment is provisioned for far more people than log into it. The instruction coming back is reasonable, obvious, and the same one every time: it is underused, so make it smaller.
You are going to have to do something with that instruction, and the honest answer is uncomfortable. Shrinking will not produce the saving being asked for. Not because the estate is efficient, and not because the request is unfair, but because of how the cost is built.
This post is the argument you need to make, and the alternative you need to bring with it. Turning up with only the first half is how people lose this conversation.
Sort your costs into two piles
Take the annual cost of the estate and sort every line into one of two piles: costs that move when seat count moves, and costs that do not.
Costs that move with seats. Per-user application licensing. Some fraction of storage. Marginal compute above the floor. Endpoint provisioning.
Costs that do not move with seats:
- The authorization program for each boundary. Assessment work is a function of the control baseline and the number of boundaries, not of how many people log in.
- The infrastructure operations team. Someone patches, backs up, monitors and maintains each cluster regardless of whether ten or a hundred people use it, and clearing that person is the expensive part.
- Minimum viable cluster redundancy. An environment with an availability requirement needs its redundancy whether or not anyone is on it.
- Manual cross-domain transfer. Appointed, trained, cleared transfer agents are a per-boundary function.
- Facility footprint and accredited space, which is bought in increments and released in increments.
- The management, security and administrative overhead attached to each boundary as an entity.
In a cleared, air-gapped estate the second pile dominates the first, and it is not close. The dominant cost is cleared labor, and cleared labor attaches to boundaries and clusters, not to users.
What that does to the shrink play
Halving your seat count moves the first pile and leaves the second one where it was.
The saving is real but modest, and it comes with a cost that does not show up on the spreadsheet: you have just formalized the estate as a small-population environment, which makes the next capability request harder to justify and hands the "why does this cost so much" question back to you in a worse form. Cost per user barely improves, because you removed users who were consuming very little.
You cannot fix a fixed-cost problem by using less of it. You can only fix it by changing what generates the fixed cost.
Which means the lever is topology
Look back at that second pile. Nearly every line in it is indexed to the same two things: how many independent authorization boundaries you operate, and how much has to sit inside the most expensive one.
That is the whole lever, and it is the reason this series spends so much time on inheritance and assessment capacity rather than on hardware.
Two moves change the second pile:
Reduce the number of independently authorized boundaries, or put something underneath them that carries controls on your behalf so each one costs less to hold.
Move work down. Most of the effort inside a high-side estate does not need to be there. Where the artifacts allow it to be separated, the high side can carry only the genuinely classified delta and a lower tier can carry the rest. That does not shrink the estate. It changes what the estate is for, which is a different operation with a very different payoff.
The second one is where most of the value is, and it has a precondition that is easy to state and hard to verify. It is the subject of post six.
The one case where shrinking is right
If your honest answer to the utilization question from the first post was demand rather than friction - the people who do not use the environment genuinely never needed it - then right-sizing is correct as a cleanup. Stop paying for seats and licenses nobody wants.
Just do not book it as the business case. It is hygiene. It will not change the shape of the estate, and the shape is what you are actually paying for.
What to say when you are told to cut
You need three sentences, and you need the third one.
Agree with the observation. The utilization figure is real and it is the right thing to have noticed. Do not defend the current state, because it is not defensible and defending it is how you become the obstacle.
Separate the observation from the remedy. The estate is fixed-cost dominated, so cutting seats removes the cheap part and leaves the expensive part untouched. Have the two piles ready, with your own lines in them. This is a five-minute conversation if you have done the sort, and an argument you lose if you have not.
Bring the alternative in the same meeting. What actually moves the number is reducing how many independent boundaries you authorize, and moving the work that never needed to be on the high side somewhere cheaper. Name what it would take to find out whether that is possible here.
That third sentence is the whole thing. "Cutting will not work" on its own sounds like a program office protecting its budget, and it will be heard that way no matter how good the arithmetic is. "Cutting will not work, and here is the thing that does, and here is the week of work that tells us whether it applies to us" is a different conversation, and it is one you are running rather than defending.
Robert Burckner is the founder of Millabs Corporation. He has served as ISSM and ISSE for legacy weapon systems and as Division Chief at the Space Warfighting Analysis Center (USSF/NRO).