AI Economics Glossary: Definitions for AI Cost, Value, and Governance
76 definitions across AI economics, TCO, ROI, FinOps, TBM, ITFM and governance. Search them or browse by letter.
A
14 terms
Absorption capacity
The capacity of an organisation to decide what to do with AI output, adopt it, change the process around it and bank the benefit. It is limited by decision slots, change slots, training and implementation capacity.
Why it matters: Value is capped by what an organisation can absorb, not by how much AI can generate.
Agentic AI
AI systems that can plan, decide, and take multi-step actions with some degree of autonomy toward a goal.
Why it matters: Agentic AI matters because more autonomous systems can create more business value, but they also increase governance, observability, and economic risk if actions scale without control.
Agentic Orchestration Cost
The cost generated by the planning, tool-calling, error recovery, and verification logic of an agentic AI system - the overhead incurred in executing a task autonomously, beyond the direct inference cost of the final output.
Why it matters: In agentic systems, orchestration cost can account for 30-60% of total inference spend and is largely invisible in standard cost reporting. Organisations that track only direct inference cost will systematically underestimate the true cost of agentic workloads.
AI Adoption
A term used in AI governance to describe user engagement with an AI capability - but applied inconsistently to mean licence assignment, feature usage, workflow integration, or behavioural change, with different evidence implications for each.
Why it matters: Adoption as a metric can mean very different things. High licence deployment with low active usage is not the same as high usage with low workflow change, which is not the same as high workflow change with low value capture. Using adoption as a proxy for value without specifying which definition is in use produces misleading performance reports.
AI Baseline
The measured pre-AI state of the process, metric, or outcome against which AI-related improvement is claimed.
Why it matters: Without a credible baseline, an AI productivity or quality claim cannot be evaluated. A baseline constructed after the AI is deployed - or estimated rather than measured - is structurally biased toward the result the sponsor needs to show.
AI FOMO (Fear of Missing Out)
The dynamic by which competitive anxiety about AI adoption leads to investment decisions that bypass normal capital discipline, justified by strategic necessity rather than demonstrable economic return.
Why it matters: AI FOMO is one of the primary structural causes of the AI Value Gap. Investments approved on competitive anxiety rather than economic evidence enter portfolios with weak proof standards, diffuse ownership, and no defined exit criteria - all of which make value realisation harder, not easier.
AI Pressure Index
A heuristic that shows whether downstream human capacity is being consumed faster than it recovers. Load at a stage is the capacity required divided by the protected capacity available. Pressure is the highest of the verify, decision and change loads. Under 0.70 is flowing, 0.70 to 0.85 is tight, above 0.85 is overloaded, and above 1.00 is unsustainable.
Why it matters: Queues form before full load, and quality can slip before a backlog is visible, so the index prompts action while there is still slack.
AI ROI
The return created by AI initiatives relative to the cost of building, operating, and governing them.
Why it matters: AI ROI matters because leaders need a credible way to judge whether AI spending is translating into measurable gains in efficiency, revenue, risk reduction, or strategic capability.
AI Sprawl
The uncontrolled proliferation of AI tools, pilots, vendors, and workflows across an organisation.
Why it matters: AI sprawl matters because duplicated tools, inconsistent standards, and weak ownership make cost, risk, and value proof harder to govern at portfolio level.
AI TCO
The total cost of owning and operating an AI capability across model access, infrastructure, data, governance, integration, and support.
Why it matters: AI TCO matters because the model price alone rarely explains the real economic burden of enterprise AI, especially once production controls and operating overhead are included.
AI Value Gap
The difference between what an organisation spends on AI and the share of that spend it can connect to a measured outcome: the complement of attribution coverage.
Why it matters: The AI Value Gap matters because organisations can scale platform cost, vendor commitments, and operational complexity before they establish enough visibility, ownership, and proof to justify continued investment.
AI value ledger
A small operating record for a material AI workflow that treats value evidence like financial evidence. The minimum record is an owner, a workflow ID, a model route, a value hypothesis, a measurement method, the evidence confidence and a scale, redesign or stop decision.
Why it matters: It joins technical telemetry, workflow metrics and business outcomes in one place, so a value claim has an owner, a baseline and an evidence level.
AI Value Management
The discipline of managing AI cost, usage, risk, adoption, quality, and realised business value across the full enterprise portfolio.
Why it matters: AI Value Management matters because AI spend can scale across SaaS, cloud, data, labour, and agentic workflows before organisations can prove which investments deserve more funding and which should be stopped.
Autonomy ladder
A five-rung scale for how much authority an AI system has in a workflow: assist, recommend, act with approval, act within policy, and self-optimise. Autonomy should follow impact, reversibility and evidence, not the novelty of the model.
Why it matters: It turns the question of how much AI can be left to do into a question about proof, with high-impact and hard-to-reverse decisions staying with humans.
C
11 terms
Capability ROI
Return created by building reusable organizational capability that improves the economics of future AI work.
Why it matters: Capability ROI matters because some AI investments do not pay back through one immediate use case but through lower marginal cost, faster delivery, and stronger governance across later deployments.
Chargeback
A mechanism for allocating and billing technology cost directly to the business unit, service owner, or consumer responsible for demand.
Why it matters: Chargeback matters because AI demand can scale quickly and expensively, and formal cost assignment can create stronger accountability for who is consuming shared capability and why.
Cloud Efficiency Posture Management (CEPM)
A discipline and tooling category focused on continuously detecting and remediating cloud waste, inefficiency, and misconfiguration.
Why it matters: CEPM matters for AI because many AI cost problems are still mediated through cloud infrastructure behaviour, especially around GPUs, clusters, and shared platform waste.
Cost Allocation
The method used to distribute shared technology cost across services, teams, products, or business units.
Why it matters: Cost allocation matters because AI costs often begin as shared platform investment and later become demand-driven; without a sensible allocation model, prioritisation and ROI discussions become distorted.
Cost Avoidance vs. Cost Reduction (AI)
Cost avoidance is the prevention of future costs that would otherwise have been incurred; cost reduction is a decrease in current expenditure. In AI business cases, these are often conflated, with cost avoidance claims presented as if they were cost reduction.
Why it matters: Cost avoidance and cost reduction have different accounting treatments, different certainty levels, and different budget implications. A CFO approving an AI investment on cost avoidance grounds should apply higher scrutiny than one approving on demonstrated cost reduction, because cost avoidance depends on a counterfactual that cannot be directly observed.
Cost per Action
The cost incurred each time an AI-enabled workflow completes a defined action such as a summary, recommendation, classification, or customer response.
Why it matters: Cost per action matters because it gives leaders a concrete way to compare AI cost against process value, service cost, or unit margin at the workflow level.
Cost per Inference
The average cost incurred each time an AI system generates one response or prediction.
Why it matters: Cost per inference matters because it gives leaders a simple but powerful way to compare AI usage behaviour against adoption and business value.
Cost per Outcome
The cost of achieving a defined business result such as a resolved case, converted lead, or completed internal workflow.
Why it matters: Cost per outcome matters because it links AI spend to business performance rather than stopping at model or workload activity.
Cost Pools
Aggregated groups of cost that are collected before being allocated, mapped, or analyzed further.
Why it matters: Cost pools matter because AI cost is often fragmented across vendors, cloud services, labor, and governance functions, and pooled views are a first step toward useful attribution.
Cost to Serve
The full cost required to deliver a capability, product, service, or workflow to its end user or customer.
Why it matters: Cost to serve matters in AI because model usage is only one layer of the economics; integration, support, governance, and platform overhead often determine whether the service is truly viable.
Counterfactual evidence level
A five-step scale, E0 to E4, for how strong the evidence is that an outcome was caused by AI. E0 is anecdote or expert opinion, E1 a before and after baseline, E2 an adjusted baseline or matched cohort, E3 a holdout group or quasi-experiment, and E4 a controlled experiment that has been independently validated.
Why it matters: An improvement after AI arrives is not proof that AI caused it. The level sets how much money or autonomy a claim can support: no monetised claim below E2.
E
1 term
F
5 terms
Fine-Tuning
Adapting a pre-trained model with additional domain data so it performs better on a narrower set of tasks. It pays when it lets a smaller, cheaper model clear the quality floor at high volume; otherwise retrieval and better prompts are usually cheaper.
Why it matters: Fine-tuning adds a one-off training cost and a recurring burden of evaluation, hosting and retraining. Whether that pays depends on volume and on what the alternative costs, not on whether the tuned model is better in a demo.
FinOps
An operating discipline for managing variable technology spend through shared accountability between engineering, finance, and business teams.
Why it matters: FinOps matters for AI because model usage, context size, orchestration choices, and infrastructure patterns can all create fast-moving cost dynamics that need continuous management rather than periodic budget review.
FinOps Scope
A formal domain of technology spend governed within the FinOps framework, such as cloud, SaaS, or AI.
Why it matters: FinOps Scope matters because AI is now recognised as a formal FinOps area, which creates a stronger basis for dedicated reporting, forecasting, and accountability.
FOCUS
The FinOps Open Cost and Usage Specification, an open standard from the FinOps Foundation that normalises billing and usage data across cloud, SaaS, data centre and other technology vendors. Version 1.4 was ratified on 4 June 2026. Support for AI and token costs is planned for version 1.5.
Why it matters: FOCUS matters because AI cost visibility depends on consistent usage and billing data, and common standards make it easier to analyze consumption across platforms and providers.
Frontier Model
A large AI model at the leading edge of current capability. Frontier models launch at premium prices that fall more slowly than those of older capability levels, so the economic question is when the extra capability is worth paying for.
Why it matters: Defaulting every task to the most capable model is one of the commonest causes of avoidable AI cost. Many tasks clear the quality floor on a smaller model at a fraction of the price.
G
3 terms
Generate, Verify, Absorb
A three-stage model of an AI-enabled workflow. Generate produces drafts, code, analysis, options and agent actions. Verify establishes that the output is right, safe and compliant to the level the risk demands. Absorb decides, adopts, changes the process and banks the benefit.
Why it matters: It shows that one stage is the binding constraint at a time, and that the constraint moves downstream when generation gets cheaper.
Governance Theater
Governance processes that generate reports, reviews, and documentation without materially changing investment decisions or operating behaviour.
Why it matters: Governance theater is expensive without being useful. It consumes management time, creates a false sense of control, and can actually slow down effective decision-making by substituting process activity for genuine scrutiny.
GPU
A graphics processing unit, the most common processor for training and running AI models. TPUs and other ASICs are specialised alternatives. For a buyer, accelerator economics turn on utilisation, commitment and who carries the cost of idle capacity.
Why it matters: Accelerator capacity is the largest cost line in self-hosted and committed AI, and it is paid whether or not it is used. A cheap hourly rate on a lightly used cluster can cost more per successful outcome than a dearer API.
I
3 terms
Inference
The process of running a trained AI model to produce an output from a prompt, query, or input signal.
Why it matters: Inference matters because most enterprise AI cost is incurred not when a model is trained, but when it is repeatedly used in production at scale.
Inference Cost
The recurring cost of generating outputs from a deployed AI system in production.
Why it matters: Inference cost matters because AI spend often scales with usage, and a workflow that appears affordable in testing can become structurally expensive once it spreads across teams, customers, or transactions.
Inference Optimisation
The practice of reducing the cost, latency, or resource intensity of AI inference without unacceptable loss of quality.
Why it matters: Inference optimisation matters because recurring production cost, not one-off model development cost, often becomes the main economic bottleneck in enterprise AI.
J
1 term
M
4 terms
Measurement Capture
The conflict of interest that arises when the team that builds or sponsors an AI system is also responsible for evaluating whether that system is creating economic value.
Why it matters: Measurement capture does not require dishonesty to produce biased results. Confirmation bias in evaluation design - choice of baseline, selection of metrics, attribution methodology, measurement timing - systematically favours positive findings when the evaluator has a stake in the outcome.
Model Hosting
The way an AI model is deployed and run in production, whether through a managed provider, dedicated service, or self-hosted infrastructure.
Why it matters: Model hosting matters because deployment choice has major consequences for cost structure, performance, governance, utilization, and operating complexity.
Model Risk
The risk of adverse outcomes arising from decisions made using incorrect, misused, or inappropriately deployed AI or analytical models.
Why it matters: Model risk is a regulated category in financial services and an increasingly important governance concept more broadly. It is also a TCO cost driver - model validation, ongoing monitoring, and compliance obligations are recurring expenses that standard AI cost models routinely underestimate.
Model Routing
The logic used to direct different requests to different models, providers, or configurations based on cost, quality, latency, or policy.
Why it matters: Model routing matters because it directly changes AI unit economics; a well-routed workflow can reserve the most expensive models for high-value tasks while using cheaper paths where appropriate.
N
1 term
O
1 term
P
2 terms
Position ledger
A dated record of the positions a publication has taken, with their current status, the last assessment and the evidence behind it, so each can be shown wrong.
Why it matters: It makes editorial claims falsifiable and shows how they have held up over time.
Prompt Efficiency
Getting the required output, at the agreed quality, with fewer tokens, cheaper models or less rework. Prompt engineering, the craft of designing prompts and context, is the means; prompt efficiency is its economic measure.
Why it matters: Prompt and context design is one of the cheapest levers on AI cost, because every token in a prompt is paid for on every call. It is also easy to overdo, because a shorter prompt that lowers quality creates review and rework that cost more than the tokens saved.
Q
1 term
R
5 terms
RAG
Retrieval-augmented generation, which supplies a model with documents retrieved at query time so its answers are grounded in the organisation's own information. It avoids retraining, but the retrieved text is billed as input tokens on every call.
Why it matters: RAG is the usual way to make a general model useful on company data. Its cost is easy to underestimate because it sits in input tokens, embeddings and search infrastructure rather than in an obvious line item.
Realised Productivity Gain
The economic value actually captured from AI-driven time savings, as distinct from the theoretical time freed by an AI capability.
Why it matters: Most AI productivity claims measure time freed, not value captured. Realised productivity gain only occurs when freed capacity is converted into lower cost, higher output, or redeployment to more valuable work - through an active management decision, not automatically.
Reasoning Tokens
Tokens a reasoning model generates internally before it gives its visible answer. They are billed at the output rate, usually not shown to the user, and often several times longer than the answer itself.
Why it matters: Reasoning tokens are one of the largest sources of error in AI cost forecasts. A workload that looks cheap from the length of its answers can cost several times more once the hidden reasoning is counted, and standard cost dashboards often show only the total.
Return on scarce capacity
Expected net AI value divided by the load a project places on the binding constraint, such as review hours, change slots, risk capacity, compute budget or executive attention.
Why it matters: Two projects with the same return on cost can differ tenfold in return on scarce capacity. It ranks work; it does not state a value.
Risk tiers: Glance, Check, Scrutinise
Three levels of human attention matched to risk. Glance is for low-stakes, reversible work, verified by automated checks and a random human sample. Check is for material but recoverable work, where a person checks the evidence pack and samples the work. Scrutinise is for high-stakes or hard-to-reverse work, with a named independent reviewer who can reject it.
Why it matters: Attention is scarce. Tiering decides how much of it each item earns, and should be set against published criteria rather than chosen by the sender.
S
6 terms
SaaS Token Aggregation
The practice of bundling and obscuring underlying token consumption within SaaS subscription pricing, creating a visibility gap between AI usage and cost.
Why it matters: SaaS token aggregation hides the relationship between token consumption and business value, making it difficult for organizations to optimize AI usage, measure productivity, or make informed procurement decisions.
Shadow AI
AI tools, models, or workflows adopted outside formal enterprise governance, budgeting, or architecture processes.
Why it matters: Shadow AI matters because it weakens visibility, creates unmanaged spend, and increases the risk that AI demand scales before accountability and proof are in place.
Shared AI Platform
A common internal platform that provides reusable AI services, tooling, guardrails, and infrastructure to multiple teams.
Why it matters: Shared AI platforms matter because they can lower marginal delivery cost and improve control, but they also create allocation, chargeback, and governance questions that are easy to ignore early on.
Showback
A mechanism for reporting technology cost back to consumers without formally charging them for it.
Why it matters: Showback matters because it creates cost visibility and accountability signals before an organization is ready to move to formal chargeback for AI services, shared platforms, or model consumption.
Sovereign AI Infrastructure
AI infrastructure operated in a way that preserves jurisdictional control over data, models, compute, and service operation.
Why it matters: Sovereign AI infrastructure matters because control requirements can materially change the economics of hosting, support, compliance, and platform design.
Stage Gate (AI Investment)
A defined decision point in an AI investment lifecycle at which continuation of investment, and release of the next tranche of resources, requires demonstrating specific evidence of value or technical readiness.
Why it matters: Stage gates are the primary mechanism for raising proof standards as investment deepens. Without them, the evidence required to scale from pilot to production is no higher than the evidence required to begin a pilot - which produces portfolios where scale decisions are made on narrative rather than on evidence.
T
11 terms
TBM Framework 2.0
An updated TBM framework used to structure how technology cost is translated into business-facing management decisions.
Why it matters: TBM Framework 2.0 matters because AI governance needs more than cost visibility; it needs a management framework that connects services, capabilities, and portfolio choices.
TBM Taxonomy 5.0
A recent evolution of the TBM taxonomy that more explicitly supports AI-related resource, solution, and consumer modelling.
Why it matters: TBM Taxonomy 5.0 matters because it gives TBM leaders a more concrete structure for representing AI types, GPU-heavy resources, model choices, and AI-specific labour.
Technology Business Management
A management discipline that connects technology cost, consumption, and investment to business capabilities, services, and outcomes.
Why it matters: TBM matters in AI because it gives leaders a way to understand whether platform cost, model usage, and shared services are supporting business capabilities in an economically coherent way.
Technology Consumers
The business units, products, functions, or capabilities that consume a technology service or solution in a TBM model.
Why it matters: Technology Consumers matter because AI cost only becomes decision-useful when leaders can see who benefits from and who drives demand for shared AI capability.
Technology Resource Towers
A TBM structure that groups technology resources into broad categories such as compute, storage, network, software, or labor.
Why it matters: Technology Resource Towers matter because AI cost often spans several resource categories at once, and leaders need a structured way to see how those layers combine.
Technology Solutions
A TBM layer used to group technology capabilities into solutions that can be mapped to business demand and service consumption.
Why it matters: Technology Solutions matter because AI often needs to be represented as a shared platform capability rather than only as infrastructure or software spend.
Token
The unit in which AI models read and write text, and the unit in which most AI usage is metered and priced. A measure of consumption, not of value.
Why it matters: Tokens set the model bill and drive capacity planning, so they are the natural unit for managing AI spend. They do not say whether anything valuable happened, so they cannot stand in for outcomes, productivity or return.
Token Productivity
The business value generated per token consumed-a measure of how efficiently AI workflows convert token consumption into meaningful outcomes.
Why it matters: Token productivity reveals that not all tokens are equal. Two workflows consuming the same token volume can deliver vastly different business value, making token productivity a critical metric for AI optimization and investment decisions.
Total cost of AI (TCA)
In the Tokenomics Foundation's value classification, the costs that cannot be written in the unit of the baseline, such as tokens, GPU hours, power, licences and the labour to build and operate the platform. Together with a value claim it would be used to determine ROI. The Foundation describes it as work in progress.
Why it matters: It marks the boundary between a value claim and a return. It also overlaps with, but is not the same as, the seven-layer AI TCO framework used on this site.
Total Cost of Ownership
The full cost of building, operating, governing, and sustaining a technology capability over time.
Why it matters: TCO matters in AI because model price alone rarely reflects the real economic burden; infrastructure, data, governance, support, and portfolio overhead often shape whether an initiative remains viable.
Training
The process of teaching an AI model patterns from data so it can perform a task. For most enterprises, training is a cost the model provider carries and recovers through its prices; the enterprise's own costs sit in inference, adaptation and operation.
Why it matters: Training dominates headlines about AI cost, but few enterprises train large models themselves. Mistaking the provider's cost structure for your own leads to business cases that miss the recurring costs of running AI in production.
U
2 terms
Unit Economics
The economics of delivering one unit of output, service, transaction, user interaction, or business outcome.
Why it matters: Unit economics matters because it translates AI from abstract platform cost into a form leaders can compare against margin, service cost, or business value at operational scale.
Utilization
The degree to which a technology asset, platform, model capacity, or workflow is actually being used relative to what it could support.
Why it matters: Utilization matters because low usage can make AI platforms look expensive even when the architecture is sound, while high usage without governance can magnify cost and control problems.
V
5 terms
Value Capture Failure
The outcome when an AI system performs as designed technically but the organisation fails to convert its technical performance into measurable economic benefit.
Why it matters: Value capture failure is distinct from technology failure. The AI worked - the value was not absorbed. This distinction matters because the remedies are entirely different: technology failure requires technical intervention; value capture failure requires operating model and management intervention.
Value claim
A defensible statement about the incremental benefit of an outcome, measured or estimated against a baseline. In the Tokenomics Foundation's value classification it is the product of the method, and it is distinct from a return or ROI, which also needs cost.
Why it matters: It keeps a benefit estimate from being presented as a return. A claim can be complete without being an ROI.
Value Realisation
The process of converting expected benefits from an investment into measurable business outcomes.
Why it matters: Value realisation matters because AI usefulness does not automatically become business impact, and leaders need proof that promised gains are actually being delivered.
Vendor Lock-in (AI)
The accumulation of switching costs - technical, contractual, and operational - that make it economically difficult or practically infeasible to change AI providers, platforms, or model sources after initial deployment.
Why it matters: AI vendor lock-in is more pervasive and harder to reverse than traditional software lock-in because it operates at multiple levels simultaneously - model API, platform architecture, data format, workflow design, and contractual terms. Organisations that do not manage lock-in risk actively will find their pricing and capability choices constrained over time.
Verification cost
The cost of establishing that AI output is right, safe and compliant to the level the risk demands: reviewer time, triage, rework, escalation and the tooling for automated checks.
Why it matters: It is a real and growing line of AI total cost of ownership that is often buried in people or governance costs, and it rises as generation gets cheaper.
Continue exploring
Follow the threads that connect AI cost, value, governance, and operating discipline.