The AI Subscription FinOps Glossary: Terms the Industry Needs

The terminology for managing enterprise AI spend is being invented in real time. In 2024, most organisations had one or two AI tools and described their costs in one word: expensive. In 2026, the average enterprise runs six to eight AI models across three to four vendors simultaneously, each with different billing models, different token counting methodologies, and different pricing change cadences. The language for governing this environment did not exist when the spending began. This glossary is our contribution to building it.

These terms were developed by PromptKing and published in full in AI FinOps: The Complete Playbook — Second Edition (June 2026). The five seat behavioural archetypes are the subject of a pending US patent application with the USPTO. All terms are offered here as working vocabulary for the field.

The Three Layers of AI Cost

AI Subscription FinOps

The discipline of managing enterprise AI spend at the licence and seat level — tracking which seats are provisioned, how they are being used relative to their plan capacity, and where recoverable budget exists through rightsizing. Distinct from infrastructure FinOps (cloud compute costs) and token analytics (API-level observability). The subscription layer is where most enterprise AI waste lives in 2026.

Infrastructure Layer

The cloud compute, GPU, and inference infrastructure costs associated with running AI workloads. Managed by cloud-native FinOps tools. Billed by the compute second, the storage gigabyte, the provisioned throughput unit.

Token Analytics Layer

The API-call-level visibility into LLM consumption — tokens per request, cost per feature, cost per customer. Relevant for teams building AI-native products on top of LLM APIs.

Subscription Seat Layer

The licence management layer — who has access to which AI tool, at what plan tier, and how much of that plan capacity they are actually consuming. The primary focus of AI Subscription FinOps.

The Five Seat Behavioural Archetypes

Patent Pending — USPTO

Every AI licence in a managed enterprise portfolio falls into one of five behavioural archetypes at any given measurement period.

Ghost Seat

A licenced AI seat generating near-zero consumption over a rolling 30-day measurement period. Licenced, billed, consuming nothing. Industry average: 18–22% of all provisioned AI seats. Highest-confidence rightsizing target — no capability trade-off in deprovisioning.

FinOps action: Deprovision or downgrade immediately. Confidence threshold: two consecutive 30-day periods below 5% of plan capacity.

Underutilised Seat

A licenced AI seat consuming a measurably low proportion of plan capacity. Active, but on the wrong plan tier. The most common waste pattern in enterprise AI portfolios — paying Max rates for Pro-level usage.

FinOps action: Downgrade recommendation with projected monthly saving. Threshold: two consecutive periods between 5% and 20% of plan capacity.

Normal Seat

A licenced AI seat consuming a healthy proportion of plan capacity — within the expected range for the plan tier, sustained over time. The reference state. No rightsizing action warranted. Document as organisational baseline.

FinOps action: No action. Monitor monthly.

Power User Seat

A licenced AI seat consuming a high proportion of plan capacity, approaching the ceiling. Well-matched to current tier — no waste — but ceiling breach risk is imminent. The goal of AI FinOps is not to turn every seat into a Power User. It is to match every seat to the plan tier that reflects its actual behaviour.

FinOps action: Monitor. Upgrade recommendation if ceiling hit in two consecutive periods. Do not downgrade.

At-Risk Seat

A licenced AI seat whose consumption pattern has changed materially from its established baseline. Defined by behavioural change, not absolute level. Two patterns, two causes: sharp decline (personnel change, workflow shift, shadow AI migration) or sharp spike (agentic cost event, misconfigured automation). Investigate before acting.

FinOps action: Investigate the cause. Do not act until cause is understood.

Seat Intelligence Metrics

Seat Utilisation Rate (SUR)

Actual AI consumption expressed as a percentage of the seat's plan capacity, measured over a rolling 30-day window. The primary metric of AI Subscription FinOps.

Interpretation: below 20% = underutilised / 20–60% = low, monitor / 60–90% = normal / above 90% = power user, ceiling risk

Recoverable Spend

The portion of an organisation's total AI budget that could be freed through rightsizing without any reduction in organisational AI capability. The primary ROI metric of an AI FinOps program.

Industry benchmark: mature programs sustain recoverable spend below 10% of total AI budget. Above 25% indicates systemic waste.

Ghost Seat Rate

The percentage of total provisioned AI seats classified as Ghost Seats in a given measurement period. Industry benchmark: 18–22% of total provisioned seats in 2026.

Plan Coverage Gap

The difference between what a seat's actual usage would cost at API list prices and what the seat holder pays through their subscription plan. Positive gap = subscription providing value. Negative gap = paying more than pay-as-you-go.

Credit Runway

In credit-based AI billing (Anthropic programmatic credits, GitHub AI Credits), the number of days remaining before a seat's monthly credit allocation is exhausted at current consumption rate. Below 14 days in the first half of a billing period signals overage risk.

Non-Pooled Credit Risk

The financial exposure created when multiple users share an automated workflow under individual credentials in a billing model where credits are per-user and do not aggregate across the team. Multiplies with team size. The primary hidden cost driver of the June 15, 2026 Anthropic programmatic billing transition.

Governance and Policy Concepts

Compliant Data Lineage (CDL)

A governance framework that defines and enforces the chain of custody for AI-generated outputs within an enterprise. CDL tracks the origin of an AI request, the model that processed it, the output produced, and any human review before the output was used in a business process. Distinct from model governance (model behaviour) and data residency (where data is processed). CDL governs the documented path of a specific AI output through an organisation's workflow.

AI Governance Score

A composite score measuring the overall compliance posture of an organisation's AI deployment — across model behaviour, data lineage, usage policy adherence, and audit completeness. An enterprise-level score reflecting organisational policies and practices, not model behaviour.

Shadow AI

AI tool usage within an organisation that occurs outside of sanctioned procurement and IT governance processes. Typically manifests as individually-expensed subscriptions, personal API keys used for business purposes, or AI tools embedded in third-party SaaS without IT approval. Invisible in vendor invoices, ungoverned by policy, and a cost and compliance risk simultaneously.

Seat Provisioning Policy

The organisational process governing how new AI licences are requested, approved, assigned, and reviewed. Absence of a seat provisioning policy is the primary cause of high ghost seat rates.

Agentic Cost Event

A discrete period during which an AI agent or automated workflow consumes tokens at a rate materially exceeding the seat's historical baseline. Primary cause of mid-month credit exhaustion in the post-June 15, 2026 Anthropic billing model. Detection requires real-time monitoring, not monthly invoice review.

The AI FinOps Category Map

Cloud-Native FinOps (Extended to AI)

Platforms originally built for cloud infrastructure cost management that have extended coverage to include AI API costs. Best suited for organisations whose primary AI cost question concerns cloud infrastructure.

LLM Observability and Token Analytics

Platforms built to instrument AI model API calls at the request level. Best suited for engineering and product teams building AI-native products.

AI Subscription FinOps

Platforms built specifically for the licence and seat layer of enterprise AI spend. Best suited for IT Directors and CFOs managing multi-vendor AI licence portfolios. The newest category — and the one most directly aligned with how most enterprises actually buy and deploy AI in 2026.

Quick Reference: Seat Archetype Decision Table

| Archetype | SUR Range | 2+ Periods? | Recommended Action | | --- | --- | --- | --- | | Ghost Seat | below 5% | Yes | Deprovision or downgrade immediately | | Underutilised | 5–20% | Yes | Downgrade to lower plan tier | | Normal | 20–80% | — | No action. Monitor monthly. | | Power User | 80–95% | — | Monitor. Upgrade if ceiling hit 2× | | At-Risk | Any (changed) | — | Investigate before acting |

A Note on Terminology Ownership

The terms Ghost Seat, Seat Utilisation Rate, At-Risk Seat, and the five-archetype seat behavioural classification system were developed by PromptKing in 2025–2026 and are the subject of a pending US patent application. The term AI Subscription FinOps was coined by PromptKing to describe the seat-level category distinct from infrastructure FinOps and token analytics. These terms are offered to the field as working vocabulary. Their use in industry discourse, analyst research, and organisational practice is encouraged. The underlying methods and systems are protected separately.

This glossary is an excerpt from Appendix D of AI FinOps: The Complete Playbook — Second Edition by Shazad Mirza (@PromptKing32). The full playbook — including 11 Board View panels for executive readers, vendor-by-vendor billing breakdowns, and a 90-day implementation plan — is available at promptking-finobs.vercel.app.

Get the full playbook →

See your portfolio's archetypes →