TVR & FinOps glossary

The discipline of connecting every technology cost (AI, cloud, data, Kubernetes, on-premise, SaaS) to the business value it produces. Operationalized through four stages: See, Attribute, Optimize, Realize. It extends FinOps across every cost domain and forward from cost visibility to value proof.

See also: the full framework

The FinOps Open Cost and Usage Specification: an open standard schema for cost and usage data that makes billing records comparable across cloud providers, data platforms, and AI vendors. Maintained by the FinOps Foundation.

An architecture in which cost records are written directly into the FOCUS schema at the moment of ingestion, with no proprietary intermediate model. The distinction matters because a translation layer adds latency, adds processing cost, and puts data fidelity behind a mechanism the customer cannot inspect. FOCUS-export-on-request is not the same property.

Invoice truth is what the vendor billed: authoritative, late, coarse. Execution truth is what actually ran: immediate, granular, unpriced. A platform holding only invoices cannot say which pipeline run caused a spike; one holding only telemetry cannot say what it cost. Joining both is a precondition for accurate attribution.

Any cost slice divided by any business metric: cost per customer, per workload, per pipeline run, per active seat. Converts an absolute spend figure into a ratio that can be judged improving or worsening. A cost number without a denominator cannot be judged at all.

A unit metric spanning multiple cost domains in one formula: for example AI-inclusive cost per customer, combining cloud, model tokens, warehouse credits, Kubernetes pods, and SaaS seats. Only computable when every domain shares one schema; point tools each produce a fraction of the answer.

Labeling every allocated cost with how it was attributed: directly tagged, inferred from usage, or distributed by a shared-service rule. A chargeback where 78% is direct and 22% is rule-distributed can be defended in a budget review; a single undifferentiated number cannot.

An allocation pipeline running as an ordered series of stages (invoice, environment, project, team) where each stage records its rule, inputs, and outputs. Any final number can be walked backwards to its source, which is what makes it audit-survivable.

Showback reports cost to a team for visibility; chargeback moves the cost onto their budget. Chargeback demands far higher attribution quality, because a number that affects someone's budget will be disputed and must survive the dispute.

A saving whose reduction has been confirmed in subsequent billing data, as distinct from identified (recommended) or applied (change made, bill not yet confirmed). Most tools report identified and let everyone assume realized. The gap is where cost programs lose executive confidence.

Total AI agent spend attributable to a set of pull requests, divided by the number of those pull requests that were merged. The signature value metric for engineering AI: it measures whether agent spend is compounding into shipped software rather than abandoned branches.

The proportion of total spend in a domain that could be attributed to specific delivered output. Reported alongside any cost-per-output metric so the denominator is explicit. A cost per merged PR computed over 40% traced coverage should be presented as exactly that.

A reference embedded in a commit linking it to the AI agent run that produced it. The join key that lets priced agent runs be attributed to merged pull requests rather than distributed across a team by proportion.

The FinOps discipline concerned with the cost behavior of AI token consumption: caching, batching, model selection, context management, and prompt efficiency. Formalized by the FinOps Foundation as a distinct practice area in 2026, acknowledging that AI spend behaves differently from infrastructure spend.

An AI workload that plans, calls tools, and iterates autonomously rather than answering a single prompt. Consumes roughly 5–30× the tokens of a chatbot interaction per task, which is why AI cost lines grow by multiples rather than percentages once agents enter production.

Reserved AI model capacity purchased for guaranteed throughput rather than billed per token. Economical above a utilization threshold and expensive below it, which makes PTU right-sizing one of the highest-value AI optimization scenarios.

Provisioned infrastructure that is reserved but unused: cluster headroom, over-requested pods, unattached volumes, idle load balancers. TVR treats it as its own attributable line rather than overhead distributed silently, because headroom nobody owns is headroom nobody reduces.

NVIDIA Multi-Instance GPU: dividing one physical GPU into isolated instances. Matching workload requirements to partition size is a primary GPU cost lever; an inference job on a full A100 when a 1g.10gb slice would serve is one of the most expensive common mistakes in AI infrastructure.

Optimizing Kubernetes across GPU, CPU, memory, storage, and network rather than CPU alone. In an AI-era cluster, CPU is frequently the cheapest resource on the node; a tool that rightsizes only CPU is optimizing the dimension that matters least.

An optimization workflow where a recommendation becomes a pull request against the infrastructure-as-code repository that owns the resource, reviewed and approved by the humans who own the service, then applied by the same Terraform run as everything else. Nothing changes silently; the audit trail is the git history.

A deployment model where the entire platform runs inside the customer's own cloud account or data center, so no cost, usage, telemetry, or workload data leaves the customer perimeter. The vendor is therefore a software provider rather than a data processor, a distinction that materially shortens vendor risk assessment under FCA, PRA, MAS, DORA, and FedRAMP.

See also: Trust Center

A vendor pricing model charging a proportion of the customer's cloud bill. Structurally misaligned with cost reduction, since vendor revenue rises with customer waste. Scales punitively as AI adoption multiplies spend without multiplying the tool's workload.

See also: how DigiUsher is priced

Microsoft Azure Consumption Commitment: a negotiated multi-year spend commitment. Eligible Azure Marketplace software purchases count toward drawing it down, which turns a tooling purchase into commitment progress rather than incremental budget.

AWS Enterprise Discount Program: committed spend in exchange for discounted rates. Eligible AWS Marketplace purchases draw down the commitment, so software bought through marketplace consumes budget already committed.

Consulting Partner Private Offer: a marketplace mechanism letting a system integrator resell a software license to an end customer through the cloud marketplace. The transaction draws down the customer's commitment while the partner bills its own implementation services.

Purchased licenses or SaaS seats that are provisioned but not actively used. Surfaced by allocating SaaS cost against real seat activity rather than seat count: the difference between what procurement bought and what the organization uses.

An open protocol letting AI assistants query external systems directly. An MCP-native cost platform allows Claude, Copilot, Gemini, or an in-house LLM to answer cost and value questions conversationally, under the same role-based access control as its dashboards.

A configuration disabling all individual-level attribution views organization-wide. Used to satisfy works-council agreements and employee-monitoring restrictions in EU jurisdictions, where per-engineer cost visibility would otherwise require separate consultation.