Why the question changed
For a decade the hard question in enterprise technology finance was what are we spending? It was a hard question, and FinOps answered it. Tagging strategies, showback reports, commitment portfolios, anomaly alerts: an entire discipline grew up around making cloud spend visible and attributable, and it worked.
Three things then happened at once. Technology spend stopped being mostly cloud: data platforms, SaaS subscriptions, and on-premise estates carried costs that never entered the cloud cost tool. AI arrived and introduced a cost line that grows by multiples rather than percentages, with agentic workloads consuming 5–30× the tokens of a chatbot per task. And boards, having approved several years of large technology investment, began asking a different question.
Not "what are we spending?" but "what did the spend produce, who owns it, and is it compounding into outcomes?"
That question cannot be answered by a cost number alone. A cost number without a value denominator is an invitation to cut the wrong thing. Spend fell 8%: good or bad? Nobody can say, because the sentence has no denominator. Cost per active customer fell 8% while customers grew 20% is a different sentence entirely, and it is the sentence an executive can act on.
The four stages
Each stage produces standalone value, and each depends on the one before it being honest. The sequence matters: attribution built on incomplete visibility produces confident nonsense, and optimization without attribution cuts whatever is easiest to cut rather than whatever is least valuable.
Stage 1: See
Land every cost record from every domain into one schema at the moment of ingestion. For TVR that schema is FOCUS, the FinOps Open Cost and Usage Specification, and "at ingestion" is the load-bearing phrase. Translating vendor formats into a proprietary internal model and then exporting to FOCUS on request is a different architecture with different properties: added latency, added processing cost, and a dataset whose fidelity depends on a translation layer you cannot inspect.
Seeing properly also means joining two kinds of truth that normally live apart. Invoice truth is what the vendor billed: authoritative, late, and coarse. Execution truth is what actually ran: immediate, granular, and unpriced. A platform holding only invoices cannot tell you which pipeline run caused the spike; one holding only telemetry cannot tell you what it cost. TVR requires both in the same query.
Stage 2: Attribute
Assign every cost to an owner and an output, and be honest about how confident you are. Attribution is where most cost programs quietly fail, because the failure is invisible: a chargeback report that looks complete is indistinguishable from one that is complete, right up until an engineering leader disputes it in a quarterly review and the whole practice loses credibility.
Two properties make attribution survive that meeting. The first is sequencing: allocation runs as an ordered pipeline (invoice, then environment, then project, then team) where each stage records its rule, inputs and outputs, so any number can be walked backwards to its source. The second is quality grading: every allocated cost carries a grade describing how it was attributed, whether directly tagged, inferred from usage, or distributed by a shared-service rule. A CFO who can see that 78% of a chargeback is directly attributed and 22% is rule-distributed can defend it. A CFO handed a single undifferentiated number cannot.
Stage 3: Optimize
Remove waste through changes that are governed, reviewable, and reversible. The industry has a recognized failure mode here: recommendation backlogs that nobody applies. Tools generate hundreds of findings, engineering teams do not trust them enough to act, and the savings stay theoretical forever.
The fix is closing the loop through the change process teams already trust. In a governed workflow automation, a recommendation becomes a pull request against the repository that owns the resource: raised automatically, carrying its evidence, reviewed by the humans who own the service, and applied by the same Terraform run that applies everything else. Nothing changes silently. The audit trail is the git history, which is exactly where auditors already look.
Stage 4: Realize
Express cost as unit economics, and verify that savings actually landed. This is the stage that distinguishes TVR from cost management, and it has two halves.
The first is the unit metric: any cost slice divided by any business metric. Each domain has a signature form: cost per workload for cloud, cost per merged pull request for engineering AI, cost per pipeline run for data platforms, cost per service for Kubernetes, cost per active seat for SaaS. Because every domain lives in one schema, these metrics compose: AI-inclusive cost per customer can include the warehouse credits, the model tokens, the Kubernetes pods, and the SaaS seats that serve that customer in a single formula. Point tools each compute their own fraction; only a platform holding the whole estate can compute the whole number.
The second is savings verification. Every optimization moves through three states: identified, applied, and verified-realized, where the third state means the subsequent billing data confirms the reduction. Most tools report the first state and let everyone assume the third. The gap between them is where FinOps programs lose executive confidence, and closing it is what turns a cost practice into a value practice.
TVR and FinOps are not competitors
TVR does not replace FinOps and does not try to. FinOps built the practice, the vocabulary, the personas, and the FOCUS specification that TVR depends on entirely. The FinOps Foundation's own trajectory points the same way: scope has widened past public cloud, and the Tokenomics discipline formalized in 2026 acknowledges that AI spend behaves differently enough to need its own treatment.
- FinOps asks: is this cost visible, attributed, and optimized?
- TVR asks: what value did this cost produce, is that value growing faster than the cost, and can we prove both to a board?
A mature FinOps practice is the prerequisite for TVR, not an alternative to it. Organizations that skip straight to value metrics without solid attribution produce unit economics nobody believes.
A maturity model that maps to the stages
The FinOps Foundation's Crawl–Walk–Run progression translates cleanly onto TVR, which is useful because most enterprises are at different maturity levels in different domains simultaneously: running in cloud, crawling in AI.
- Crawl: one domain visible in FOCUS, ownership assigned at business-unit level, a handful of optimization scenarios running, one unit metric published to leadership.
- Walk: three or more domains in one schema, auditable sequenced chargeback with quality grading, optimization flowing through the governed pipeline, unit metrics per product line, savings verified rather than estimated.
- Run: every domain including on-premise and SaaS, attribution trusted enough to drive budgets, composite unit metrics such as AI-inclusive cost per customer in the board pack, forecasts using external business signals, and value questions answered conversationally by AI assistants querying the platform directly.
How to adopt it
Start with the domain where the gap between spend and understanding is widest. For most enterprises in 2026 that is AI: spend growing fastest, attribution weakest, executive scrutiny highest. Instrument one stage at a time, and do not move to attribution until visibility is complete for that domain.
Two things are worth insisting on regardless of vendor. First, that the cost dataset is FOCUS-conformant and exportable, so the practice you build is portable and does not become a hostage to the tool. Second, that every savings claim distinguishes identified from verified-realized, because the day someone in finance discovers the difference is the day your program's credibility is decided.