Why most IT unit costs never change a decision

In brief

Many TBM programs publish dozens of unit costs that rarely change a decision. Why that happens, what makes a unit cost useful, and a test to prune the list.

Steve Elliott

Steve Elliott · · 7 min read


On this page
  1. What does the TBM Council say unit costs are for?
  2. Why do most IT unit costs never change a decision?
  3. What makes a unit cost useful?
  4. An illustrative example: cost per VM vs cost per order
  5. How do you prune a unit-cost catalog?
  6. Where do AI unit costs fit?
  7. Sources

In our view, most IT unit costs never change a decision because they are priced in units only IT recognizes, and nobody who reads them can move them. Cost per server, per GB and per ticket are real numbers, but a business leader can't buy fewer of them, and the tower owner who can often isn't asked to.

A unit cost earns its place when it is tied to a volume the consumer recognizes, owned by someone who can change it, stable enough to trend and reported next to a quality measure. In our view, many catalogs would shrink sharply under that test. This post sets out the argument, a worked example with made-up numbers, and a five-question test for pruning.

What does the TBM Council say unit costs are for?

The TBM Council's KPIs and metrics page defines a unit cost as total cost of service divided by the number of units consumed. Its examples are resource units: cost per VM, per GB of storage, per network port. It says the measure enables service-based costing and supports cost recovery, benchmarking and operational decision-making, and that unit costs reveal whether services are cost-effective, overprovisioned or subsidized. The page's framing is that "these metrics don't just report results—they shape decisions."

Two details on that page matter for this argument. First, it places unit costs in its second maturity phase, where the emphasis shifts to accountability at the service owner level. Second, it treats allocation coverage as the earlier step, a way to show progress before granular metrics such as unit cost or TCO exist. The Council's CFO guide lists unit costs and TCO as the "decision-ready intelligence" a TBM model produces, and its CIO guide says those unit costs are fully burdened, reflecting the actual resource mix behind them.

We agree with all of this. Our point is narrower: publishing a unit cost is not the same as wiring it to an owner and a decision, and many programs stop at publishing.

Why do most IT unit costs never change a decision?

These are our observations, not survey findings. Usually more than one applies.

  1. The wrong unit. An earlier version of the FinOps Foundation's unit cost capability gives the clearest example. If you store the same data in fewer GB, it notes, "your cost-per-GB does not change even though your storage costs have decreased." A cost per stored item would show the saving, which in its example is about 30%. Many resource unit costs reward or punish the wrong thing in the same way. A service desk that moves password resets to self-service can see cost per ticket rise, because the cheap tickets left.
  2. No business volume behind it. Cost per VM has a denominator IT controls. The business consumes orders, claims, shipments and active customers. If the unit cost doesn't connect to one of those, no one outside IT can plan with it.
  3. No owner. If no named person can change either the cost or the volume, the number gets reviewed and filed. The Council's own framing ties unit costs to service owner accountability. A unit cost without a service owner is a report, not a lever.
  4. Unstable allocations. If allocation rules change every quarter, the unit cost moves for reasons that have nothing to do with efficiency. After two unexplained swings, readers soon stop trusting the trend, and a unit cost that can't be trended has little left to offer.
  5. No benchmark or target. A number with nothing to compare it to invites no action. Is $1,320 per VM a year good? Without a target, a prior-year figure or a peer comparison, nobody can say.

What makes a unit cost useful?

We think a useful unit cost has four properties. The FinOps Foundation's unit economics capability points the same way: it describes unit economics as "tying cloud costs more tightly to business outcomes," and says engineers can more easily start with resource efficiency metrics, and that business unit metrics such as cost per tenant or per transaction indicate higher maturity.

  • A unit the consumer recognizes. Cost per order, per claim, per employee onboarded, per active customer. The test is whether the business leader paying for it already counts that thing.
  • An owner who can change it. Someone must be able to move the numerator (cost) or the denominator (volume). Ideally both sides have a name: a service owner for cost and a business owner for demand.
  • Stable enough to trend. Lock the method for a year. When allocation rules must change, restate the prior periods so the trend still means something.
  • Paired with quality. A unit cost alone can be cut by degrading the service. Report it beside the measure that would reveal that: availability, error rate, resolution time, customer satisfaction. The Council's KPI page describes solution health scores that combine reliability, performance and cost-effectiveness, among other measures, which is the same instinct.

None of this makes resource unit costs worthless. Cost per VM or per GB is a sound diagnostic for a tower owner setting rates or comparing on-premises cost with cloud. The problem comes when they are put in front of business leaders as if those leaders could act on them.

Resource unit costBusiness-facing unit cost
ExampleCost per VM, per GB, per ticketCost per order, per claim, per active customer
Who recognizes the unitITThe consuming business
Who can change itTower or service ownerService owner (cost) and business owner (demand)
Typical decisionRate setting, sourcing, capacityBudget, pricing, product investment
Best audienceIT leadership, engineeringBusiness leaders, CFO

An illustrative example: cost per VM vs cost per order

The numbers below are made up to show the mechanics. They don't come from any company.

Infrastructure unit cost. A compute tower costs $7.2 million a year and runs 6,000 VMs, so cost per VM is $1,200. The infrastructure team consolidates workloads. Next year the tower costs $6.6 million and runs 5,000 VMs. Total cost fell by $600,000, but cost per VM rose to $1,320. Reported on its own, the unit cost says the team got worse. It also gives no business leader anything to act on, because none of them orders VMs.

Business-facing unit cost. An order-to-cash service, fully burdened with its share of applications, infrastructure and support labor, costs $9.6 million a year and processes 4 million orders: $2.40 per order. Next year orders grow to 4.6 million and cost rises to $10.1 million. Cost per order falls to about $2.20, while order processing success holds at 99.6%.

Year 1Year 2Change
Compute tower cost$7.2M$6.6M-8%
VMs6,0005,000-17%
Cost per VM$1,200$1,320+10%
Order-to-cash cost$9.6M$10.1M+5%
Orders processed4.0M4.6M+15%
Cost per order$2.40about $2.20-8.5%

The second number supports decisions the first can't. A sales operations leader can use it to price a new channel. A CFO can see that the service scales: volume grew 15% while cost grew 5%. The service owner has a target to work toward. If that compute tower serves order-to-cash, its consolidation savings show up inside the cost per order, where they belong.

How do you prune a unit-cost catalog?

Run each published unit cost through five questions. We suggest keeping it in business-facing reports only if the answer to all five is yes.

  1. Did it inform a named decision in the last 12 months, or is one scheduled? A renewal, a rate change, a budget, a price. "It's in the monthly pack" doesn't count.
  2. Would the consumer recognize the unit without a definition?
  3. Is there a named owner who can change the cost or the volume?
  4. Has the method held steady for four quarters, or been restated?
  5. Is there a target or comparison, and a quality measure beside it?

Then sort the results three ways:

  • Keep. Passes all five. Give it a target and put it in front of its owner before the budget or renewal it informs.
  • Demote to diagnostic. Fails question 2 but passes the rest. Keep it in the tower or engineering view, where cost per GB is useful, and take it out of business reports.
  • Retire. Fails questions 1 and 3. If nobody used it and nobody can move it, stop calculating it.

Anything else gets a fix and a second look next quarter: name the owner, set the target, restate the history or add the quality measure.

The Council's KPI page lists dozens of candidate measures across three maturity phases. In our view it is better read as a menu than a checklist: pick the few IT KPIs your decisions need. This is the same logic as restarting a stalled transparency program: work back from the decision, not forward from the catalog.

Where do AI unit costs fit?

We expect AI to add a new row to many catalogs, and the same test applies. Cost per token is a resource unit, like cost per GB. The business doesn't consume tokens. It consumes resolved tickets, processed documents and drafted contracts. The TBM Council's Beyond Tokenomics paper argues that "if you are only focused on token costs you are missing the big picture," because AI economics also span labor, infrastructure, data, software and process change.

We think the useful AI unit cost is cost per completed task, fully burdened, paired with an accuracy or acceptance rate. We worked through the token-versus-labor split in AI initiative costs: tokens vs labor.

A unit cost is only as useful as the decision it reaches. That is the case for an operating model in which IT service costing sits next to the services, owners and funding choices it is meant to inform.

Sources

Frequently asked questions

Which IT unit costs should we report to the business?

In our view, only those priced in a unit the business already counts, such as cost per order, per claim or per active customer. Each should have a named owner who can change it, a method stable enough to trend, a target or comparison and a quality measure beside it. Resource unit costs such as cost per VM are better kept in IT's own reporting.

Are cost per VM and cost per GB useless?

No. They are sound diagnostics for tower owners setting rates, sizing capacity or comparing on-premises with cloud cost, and the TBM Council lists them as unit cost examples. They mislead when shown to business leaders who can't act on them, or when consolidation raises the unit cost even as total cost falls.

How often should unit cost methods change?

As rarely as possible. We suggest locking allocation rules for at least a year so the unit cost can be trended. When a rule must change, restate prior periods on the new basis. A unit cost that moves because the method changed, not the service, quickly loses the trust of the people meant to use it.

What is the difference between a unit cost and unit economics?

A unit cost is the cost of delivering one unit of a service, such as one VM or one order. Unit economics, as the FinOps Foundation describes it, connects cloud spend to business value, such as cost per transaction or per customer. In our view, the business-facing unit costs worth publishing are the ones that work as unit economics.

Free playbooks

In-depth guides for product and operations leaders.

Browse the playbooks
Steve Elliott

Steve Elliott

Published November 22, 2026

More on TBM & IT Financial Management

Cost pools, towers, showback, unit cost.