Solveo ConsultingSolveo Consulting

Glossary & Lexicon

Overview

The map of disciplines

Build

Operate

Steer value

These disciplines work in synergy. A FinOps practice finds its anchoring through the CCoE, while the CCoE gains its economic value dimension through FinOps. That mutual reinforcement extends across every environment, from AI governance to carbon steering (GreenOps).

CCoE: Cloud Center of Excellence

Definition

A cross-functional body that defines, arbitrates and enforces Cloud usage rules across the organisation, and equips teams to follow them willingly.

In practice

The CCoE is where an organisation decides what is permitted by default, what requires a waiver, and who owns each trade-off. Its maturity shows in its ability to get standards adopted, beyond simply publishing them. Extended to AI usage, it becomes a Cloud & AI Center of Excellence.

What it covers

  • Target operating model and explicit RACI: who decides, who executes, who is consulted on cost, security and architecture
  • Landing zones, guardrails and policy as code: compliance by default rather than retrospective control
  • Service catalogue and reusable architecture patterns
  • Management of exceptions and waivers, with a stated duration and a named owner
  • Enablement: training, documentation and support for product teams
  • Native integration of FinOps, security, sovereignty and GreenOps considerations from design onwards (shift left)

Steering indicators

  • Share of workloads deployed into a compliant landing zone
  • Standards compliance rate, number of active waivers and their age
  • Average lead time to provision a compliant environment
  • Service catalogue adoption rate among product teams
  • Share of architecture decisions supported by a financial estimate

Key distinctions

  • The CCoE designs the rules and equips teams, execution remains owned by product teams.
  • Its value comes from being cross-functional: it brings together architecture, security, finance and the business.
  • It is a permanent function, with a lasting mandate and budget.
  • Its authority to arbitrate flows directly from executive sponsorship.

Who is concerned

  • CIO/CTO
  • Cloud Leaders
  • Architecture
  • Security
  • Product Owners

Related terms

AI Governance

Definition

A decision, control and accountability framework applied to artificial intelligence usage: which use cases are authorised, under which conditions, at what cost, and with what level of evidence.

In practice

Generative AI has reversed the adoption chain: usage arrives through the business, often ahead of any architecture or procurement decision. AI governance therefore starts by making visible what already exists (shadow AI), then setting risk-proportionate rules that keep usage inside a controlled perimeter.

What it covers

  • Inventory of use cases, classified by risk level and criticality
  • Usage policy: permitted data, permitted models, human oversight obligations
  • Build / buy / API trade-offs, and the choice between RAG, fine-tuning and prompt engineering based on full cost and the level of control required
  • Governance of input and output data: confidentiality, intellectual property, retention, traceability
  • AI-specific security: prompt injection, data exfiltration, access management for models and agents
  • Quality and drift evaluation: test sets, regression measurement across model versions
  • Compliance: GDPR, the European AI Act, ISO/IEC 42001, NIST AI RMF
  • Economic steering of usage, directly interfacing with the FinOps practice (see Tokenomics)

Steering indicators

  • Share of AI use cases inventoried and attached to a business owner
  • Full cost per use case and per active user
  • Measured value: time saved, resolution rate, actual adoption after 90 days
  • Share of use cases classified against regulatory obligations
  • Number of incidents related to data or answer quality

Key distinctions

  • A policy earns its value when it comes with control mechanisms and named owners.
  • Well designed, governance accelerates go-live by resolving legal uncertainty upstream.
  • Compliance sets the frame, most day-to-day trade-offs remain economic and technical.

Who is concerned

  • CIO/CTO
  • CFO
  • Legal & compliance
  • CISO
  • AI Leaders
  • Business teams

Related terms

DevOps

Definition

A set of practices and organisational principles that shorten the cycle between a product decision and its reliable release, by bringing development and operations together around shared responsibility.

In practice

For an executive, DevOps is first a matter of speed and risk, measurable through four stable indicators (DORA). It is also a financial matter: in the Cloud, every automated deployment is a purchasing decision. A high-performing DevOps chain, fitted with economic guardrails, industrialises spend control as much as delivery.

What it covers

  • Continuous integration and delivery (CI/CD), automated testing
  • Infrastructure as Code: infrastructure becomes a versioned, reviewable and controllable artefact
  • Observability: logs, metrics, traces, and correlation with cost
  • Internal platforms and platform engineering: providing paved roads that make good practice the natural choice
  • Variants: DevSecOps (embedded security), GitOps (Git as the source of truth), MLOps and LLMOps (model lifecycle)

Steering indicators

  • Deployment frequency and lead time for changes
  • Change failure rate and time to restore service (MTTR)
  • Share of infrastructure managed as code
  • Share of cost and security controls embedded in the pipeline (policy as code)

Key distinctions

  • Tooling serves the approach, performance comes from team practices and shared responsibility.
  • It is an organisational model shared between development and operations, carried by the teams themselves.
  • Agile concerns how the product is steered, DevOps how reliably it is delivered: the two reinforce each other.

Who is concerned

  • CTO
  • Engineering leadership
  • Product Owners
  • Architecture
  • FinOps Lead

Related terms

AIOps

Definition

The application of data analysis and machine learning to IT operations, in order to detect, explain and correct anomalies faster than threshold-based monitoring allows.

In practice

AIOps answers a volume problem: distributed environments emit far more signals than rule-based human monitoring can absorb. Its value is judged on the reduction in alert noise and diagnosis time, beyond the sophistication of the models involved. Applied to billing data, the same mechanics feed the FinOps practice directly.

What it covers

  • Alert correlation and deduplication, noise reduction
  • Anomaly detection on technical metrics as well as consumption and cost data
  • Assisted root cause analysis and automatic enrichment of incident context
  • Automated remediation for known scenarios, capacity and load forecasting

Steering indicators

  • Reduction in alert volume and false positive rate
  • Mean time to detect and to restore
  • Share of incidents handled through automated remediation
  • Quality of capacity and cost forecasts (actual versus forecast variance)

Key distinctions

  • Its performance rests first on data quality and on a reliable configuration repository.
  • It is an analysis layer that builds on existing observability and increases its value.
  • Automated remediation calls for its own governance and a regular review of its rules.

Who is concerned

  • CTO
  • Operations
  • SRE
  • FinOps Lead

Related terms

FinOps

Definition

The discipline of managing the value of variable technology spend, bringing finance, technology and business teams together around shared accountability for consumption trade-offs.

In practice

FinOps is about value before it is about spend. Reducing an invoice is a frequent outcome, and it gains from sitting within a broader objective: knowing what a service costs, who benefits from it, what value it produces, and being able to arbitrate between performance, speed, risk and cost. Mature organisations look for a governance model that endures, survives reorganisations and absorbs new scopes. The FinOps Foundation framework structures the practice in three iterative phases, Inform, Optimise, Operate, broken down into domains and capabilities.

What it covers: spend scopes

  • Public Cloud: IaaS, PaaS, commitments and contractual discounts
  • Artificial Intelligence & GenAI: inference, training and agent costs (see the dedicated entry)
  • Data platforms: Snowflake, Databricks, BigQuery and consumption-based models
  • SaaS and licensing: usage rights, framework agreements, software asset management
  • On-premise and data centre: depreciation, capacity, unit cost comparable to Cloud

What it covers: capabilities

  • Visibility and financial data quality: tagging taxonomy, shared cost allocation, showback and chargeback
  • Unit economics: cost per transaction, per customer, per environment, the basis of a sound trade-off
  • Budgets, forecasts and drift detection, with thresholds designed jointly with application owners
  • Usage and rate optimisation: rightsizing, waste elimination, commitments (Reserved Instances, Savings Plans, CUDs), architectures matched to the workload profile
  • Evidence-based support for vendor negotiation: consumption assumptions, commitment scenarios, market comparisons
  • Organisational anchoring: operating model, RACI, steering bodies, team enablement

Steering indicators

  • Allocation coverage: share of spend attributable to a named owner
  • Business unit cost and its trend at constant volume
  • Forecast accuracy (budget versus actual variance)
  • Commitment coverage and utilisation rates, effective savings rate
  • Waste identified and waste actually remediated
  • Maturity level of the practice and number of teams genuinely engaged

Key distinctions

  • Cost reduction is a frequent outcome, and it should remain one objective among several: some trade-offs lead to deliberately spending more.
  • A visualisation platform informs the decision, accountability and arbitration remain with the organisation.
  • Savings endure through the operating model: that is what sustains them beyond two or three quarters.
  • Joint commitment from Finance and Procurement makes the practice fully operative.

Who is concerned

  • CFO
  • CIO/CTO
  • Procurement
  • Product Owners
  • Cloud Leaders
  • Management control

Related terms

Tokenomics: FinOps applied to AI

Definition

The discipline of measuring, allocating and arbitrating the costs of generative AI usage, whose base economic unit is the token consumed at model input and output.

In practice

Cloud introduced variable spend driven by architecture. Generative AI introduces variable spend driven by usage, therefore by user behaviour and prompt design, two variables that belong as much to product and business teams as to the IT function. The cost of an assistant follows the number of conversations, their length and the model selected. The relationship between FinOps and AI runs both ways: governing the cost of AI, and using AI to govern costs.

What it covers: FinOps for AI

  • Cost model: input and output tokens, context window, prompt caching, batch processing
  • Model arbitration: route simple requests to lower-cost models, reserve advanced models for high-value cases
  • Architecture comparison: RAG, fine-tuning, agents, considering full cost, latency and maintainability
  • Associated infrastructure costs: GPU hours, provisioned throughput or pay-as-you-go, vector storage
  • Allocation by use case, team and user, the condition for credible showback
  • Agent costs: chained calls multiply consumption in counter-intuitive ways and benefit from being capped

What it covers: AI in service of FinOps

  • Consumption anomaly detection and automatic qualification of variances
  • Improved accuracy of budget forecasts
  • Automation of reporting and of recurring written analysis
  • Optimisation recommendations, contextualised and prioritised by impact

Steering indicators

  • Cost per request, per conversation, per active user and per use case
  • Cost per unit of business value: ticket resolved, document processed, line of code accepted
  • Input to output token ratio and cache hit rate
  • Share of AI spend allocated to a named owner
  • Marginal cost of scaling: projection at ten times current volume

Key distinctions

  • Its cost drivers are primarily product and behavioural, which sets it apart from a conventional Cloud invoice line.
  • Every organisation is concerned: office productivity usage drifts as quickly as technical usage.
  • Product design weighs more on the invoice than the model's unit price.

Who is concerned

  • CFO
  • CIO/CTO
  • AI Leaders
  • Product Owners
  • Procurement

Related terms

GreenOps

Definition

The operational practice of measuring, arbitrating and continuously reducing the environmental footprint of digital workloads, with the steering rigour FinOps applies to cost.

In practice

GreenOps is to Green IT what FinOps is to management control: a continuous decision mechanism, wired into the teams that design and operate. Its difficulty is above all methodological: emissions data published by providers remains heterogeneous, partial and hard to compare. An important corollary for executives: cost and carbon often converge (an idle resource both costs and emits) and sometimes diverge (a low-carbon region can be more expensive). Steering therefore rests on an explicit trade-off.

What it covers

  • Measurement and estimation of emissions by service, application and team, with a stated and documented methodology
  • Elimination of idle resources and rightsizing, the first lever shared with FinOps
  • Region selection and workload scheduling according to grid carbon intensity
  • Eco-design of architectures and digital services, restraint in the data retained
  • AI-specific trade-offs: model size, retraining frequency, inference footprint at scale
  • Training of technical teams and integration of criteria into architecture reviews

Steering indicators

  • Estimated emissions per application and per business unit of work (gCO2e per transaction)
  • Software Carbon Intensity (Green Software Foundation specification)
  • Share of workloads hosted in low-carbon-intensity regions
  • Average utilisation rate of provisioned resources
  • Efficiency of owned facilities: PUE, WUE, CUE

Key distinctions

  • The robustness of the figures rests on an explicit, documented and auditable methodology.
  • The overlap with FinOps is partial: some environmental trade-offs carry an accepted cost.
  • Beyond the regulatory frame, GreenOps is an architecture and design lever.

Who is concerned

  • CIO/CTO
  • Sustainability leadership
  • CFO
  • Architecture
  • Procurement

Related terms

Green IT: Sustainable digital

Definition

An overall approach to reducing the environmental footprint of the information system across its whole lifecycle, from devices to hosting, including service design.

In practice

Green IT sets the frame, GreenOps carries out continuous execution on workloads. Two distinctions structure the subject: Green IT (reducing the impact of digital) and IT for Green (using digital to reduce the impact of the rest of the business). Above all, the predominance of hardware: in most organisations, device manufacturing weighs more than hosting. Extending the life of a device fleet often produces more effect than the sum of infrastructure optimisations.

What it covers

  • Hardware lifecycle: responsible purchasing, extended service life, reuse, end-of-life channels
  • Manufacturing footprint (embodied carbon) and lifecycle analysis
  • Eco-design of digital services and the associated reference frameworks (RGESN, GR491)
  • Data restraint: retention policies, reduced duplication, appropriate archiving
  • Procurement: environmental criteria in tenders and supplier clauses
  • Non-financial reporting: CSRD and the ESRS E1 standard, the IT contribution to the carbon balance, scope 3 in particular

Steering indicators

  • IT carbon balance by category: devices, network, hosting, usage
  • Average device lifespan and reuse rate
  • Share of projects that went through an eco-design review
  • Volume of stored data relative to its actual use
  • Reliability and auditability of non-financial reporting data

Key distinctions

  • The footprint spreads across devices, network, hosting and usage: hosting accounts for one part of it.
  • The framework delivers its value when the indicators feed purchasing and design decisions.
  • Green IT sets the frame and the objectives, GreenOps executes them day to day.

Who is concerned

  • CIO
  • Sustainability leadership
  • CFO
  • Procurement
  • Executive committee

Related terms

A→Z lexicon

Cross-cutting lexicon

Attribution of a technology cost to the entity that consumes it: team, product, cost centre or use case.
FinOps & Budgets IT
Steering principle by which each iteration of the FinOps cycle refines thresholds, allocations and optimisation levers as maturity progresses.
FinOps & Budgets IT
A significant, unanticipated deviation from the expected consumption profile.
CAPEX denotes capital expenditure. A purchased IT asset, such as a server, sits in CAPEX and is depreciated over several years. This accounting regime shapes the funding of Private Cloud and owned infrastructure.
OPEX denotes operating expenditure. A Public Cloud resource consumed on demand sits in OPEX and is expensed as it is used. This accounting regime shapes the funding of Public Cloud and on-demand consumed services.
An operational activity of the FinOps reference framework, attached to a domain and primarily worked within one of the cycle phases.
FinOps & Budgets IT
Emissions generated by the production of equipment, amortised over its operating lifetime.
GreenOps & Green IT
A mechanism by which the technology cost is actually charged to the budget of the consuming entity.
Private Cloud
A US federal statute (Clarifying Lawful Overseas Use of Data Act) allowing United States authorities to access, under defined conditions, data held by US providers, including outside US territory. A central factor in European digital sovereignty trade-offs.
An open model for estimating Cloud energy and emissions, providing average power coefficients per vCPU, per graphics processor, per storage and per network unit. These coefficients are indicative values: they vary widely with processor generation and manufacturer, and should be refined by instance family on a real engagement.
GreenOps & Green IT
A Cloud offering operated under legal and/or technological sovereignty criteria (immunity from extraterritorial laws, hosting and operation by national or European entities). It differs from traditional Public Cloud through its guarantees against extraterritorial access.
A repository listing the configuration items of an information system (servers, applications, dependencies) and their relationships. An indispensable foundation, alongside ITAM, for building a reliable cost allocation model in Private Cloud.
A FinOps KPI measuring, on a 0 to 100 scale, the ratio between identified savings potential and total cost on a given scope (team, service, account). An efficiency score rather than a raw amount.
An indication signalling whether a dataset is final or remains subject to revision.
FOCUS Standard
A practitioner or adjacent function exploiting normalised datasets to conduct FinOps activities.
FOCUS Standard
The sum of all costs associated with a service over its life cycle: subscription, usage, support, integration and supervision.
Accumulation of conversation history resent on each call, making the later turns of a conversation mechanically more expensive.
FinOps for AI
The European sustainability reporting framework requiring the companies concerned to publish standardised and verified environmental information.
Compliance & CSRD
A pricing mechanism (Google Cloud) granting a discount in exchange for a consumption commitment over a defined term, the functional equivalent of Reserved Instances (AWS) and Savings Plans.
A continuous loop of three phases, Inform, Optimise, Operate, each mobilising precise capabilities of the framework.
FinOps & Budgets IT
The unit measuring compute capacity consumed on the Databricks platform, converted into cost at a rate per unit of time. The DBU is to Private Cloud what the SKU is to Public Cloud: a pricing benchmarking brick.
A management system centralising the inventory and tracking of the physical assets of a data centre (servers, network, power, cooling). It complements the CMDB on the physical infrastructure side.
Processing capacity reserved in advance with an AI provider, guaranteeing a latency and a rate.
FinOps for AI
Scheduling of time-tolerant workloads on the grid's lower carbon intensity windows.
GreenOps & Green IT
A platform-specific unit of account (Snowflake credit, Databricks DBU, BigQuery slot) used to measure and bill consumption. Converted into real currency at a contractually set rate, it introduces an abstraction layer between technical usage and the amount invoiced.
A grouping of capabilities sharing a common objective: understand cost and usage, quantify business value, optimise cost and usage, steer the practice.
FinOps & Budgets IT
A reporting principle documenting both the impact of the activity on the environment and the impact of environmental issues on the activity.
Compliance & CSRD
The set of usage rights acquired contractually from a provider, to be compared with the usage actually observed.
SaaS & Licensing
Integration of environmental criteria from the design of a service, across architecture, code, data and interfaces.
Eco-design
Linking technology cost to a business indicator: cost per customer, per transaction, per document processed.
FinOps & Budgets IT
Water consumed upstream of IT usage, to generate and distribute the electricity, distinct from the water consumed directly by facility cooling. It is to water what Scope 3 emissions are to carbon.
GreenOps & Green IT
The volume of water consumed relative to the energy used by the IT equipment of a facility.
GreenOps & Green IT
The ratio between the total energy consumed by a facility and the energy actually used by the IT equipment.
GreenOps & Green IT
Removal of already-consumed tool observations and compression of prior steps to contain the context volume.
FinOps for AI
Emissions from hardware manufacturing, facility construction and energy production, accounted for before any operation begins. On a decarbonised grid they dominate the life-cycle balance and shift the lever of action towards hardware utilisation rate and service life.
GreenOps & Green IT
Cost multiplication through successive delegation to sub-agents, each branch reproducing the full cost structure of the workflow.
FinOps for AI
An open specification that normalises billing and usage data across technology providers, making it comparable within a single reference model.
FOCUS Standard
The distinction between the entity that commercialises a service and the one on which the underlying resource is deployed, notably in the presence of resellers.
FOCUS Standard
A provisioned and billed resource whose usage remains nil or marginal.
A provider producing a dataset conforming to the specification: Cloud provider, SaaS publisher, AI service, data platform, internal team practising recharge.
FOCUS Standard
The discipline of inventorying, tracking and bringing into compliance the equipment and licenses held by the organisation.
SaaS & Licensing
The set of mechanisms bounding the number of iterations of an agent: turn budget, tool scoping, explicit stop condition.
FinOps for AI
The set of approaches aimed at reducing the environmental impact of digital technology, from equipment design to end of life.
GreenOps & Green IT
The practice of measuring, reducing and steering the environmental footprint of information systems, articulated with daily technical decisions.
GreenOps & Green IT
A technical rule applied automatically to keep configurations within the defined frame.
An architecture model centralising shared functions (governance, catalogue, security) in a hub, with consuming domains or teams (spokes) operating at the edge. One of the reference patterns for workload placement on data platforms.
Running an already trained model on a request. Once training is amortised over the volume served, inference concentrates most of the energy consumption of an AI service in production.
GreenOps for Artificial Intelligence
A versioned description of infrastructure, deployed automatically and reproducibly.
The quantity of emissions associated with producing a unit of electricity on a given grid, varying by location and time.
GreenOps & Green IT
The emission rate of the electricity generation called at the margin to answer one additional megawatt-hour, the relevant indicator for measuring the real effect of shifting or relocating a workload, unlike average intensity which describes the grid mix.
GreenOps & Green IT
The discipline managing the lifecycle of IT assets (hardware and software), from acquisition to decommissioning. Together with the CMDB, it provides the data foundation required for any FinOps governance on Private Cloud.
The set of line-by-line charges, normalised according to the FOCUS schema.
FOCUS Standard
A set allowing line-by-line reconciliation of cost and usage data with the invoice issued by the provider.
FOCUS Standard
A set describing the terms of a commitment: dates, remaining units, consumption, benefit tier reached.
FOCUS Standard
A preconfigured Cloud foundation, compliant by default, on which teams deploy.
Scope 2 accounting using the average carbon intensity of the grid supplying the site.
GreenOps & Green IT
Scope 2 accounting using the contractual electricity attributes acquired by the organisation.
GreenOps & Green IT
Four widely adopted software delivery performance indicators.
A mechanism retaining the computed representation of a stable prompt prefix, to charge subsequent calls at a reduced rate.
FinOps for AI
Returning a previously produced response for a semantically equivalent query, without a new call to the model.
FinOps for AI
The number of parameters a model actually engages to process a request, which can be far below the total parameter count on mixture-of-experts architectures. It is this quantity, not the headline size, that drives the energy consumed per token.
GreenOps for Artificial Intelligence
A choke point centralising calls to models, assigning a virtual key per team or use case and applying spend caps.
FinOps for AI
A spend segment on which the framework applies: product, cost centre, environment or custom breakdown.
FinOps & Budgets IT
A function engaged in the FinOps practice: practitioner, engineering, finance, leadership, procurement, product.
FinOps & Budgets IT
A stage of the FinOps work cycle: Inform, Optimise, Operate.
FinOps & Budgets IT
Applying spend limits ahead of consumption: per-session cap, per-team quota, automatic pause.
FinOps for AI
A compliance policy expressed as code and evaluated automatically.
A structuring conviction that guides how FinOps is practised, independently of the activities carried out.
FinOps & Budgets IT
Keeping equipment in operation beyond its usual amortisation period, to spread its embodied footprint over a longer period.
Eco-design
A responsibility matrix distinguishing responsible, accountable, consulted and informed.
Enriching a query with company-specific data, without retraining the model.
Removal of functional overlaps between applications subscribed separately by different teams.
SaaS & Licensing
The share of accounts actually used among open accounts, a direct indicator of savings potential.
SaaS & Licensing
A clause automatically extending a contract without explicit intervention, to be monitored against notice periods.
SaaS & Licensing
Reclaiming inactive or underused licenses for reassignment.
SaaS & Licensing
An adjustment billed when observed usage exceeds the rights acquired at a contractual milestone.
SaaS & Licensing
Non-guaranteed compute capacity, sold at a heavily reduced rate and reclaimable by the provider at short notice. Suited to eviction-tolerant workloads (batch processing, distributed compute, testing), it improves the overall utilisation rate of the estate.
Optimisation and eco-design
The ratio between the load actually running on a node and the capacity it reserves. Low density leaves reserved resources unused, drawing energy and carrying the whole of their manufacturing emissions.
Optimisation and eco-design
Reducing the numerical precision of a machine learning model's weights, which lowers memory footprint and inference cost at the price of a quality loss to be measured on real use cases.
Optimisation and eco-design
Putting hardware back into service for an equivalent use, internally or through external refurbishment, ahead of any recycling. Reuse avoids the manufacturing emissions of a new device, whereas recycling recovers only a fraction of the materials.
Optimisation and eco-design
The French general reference framework for the eco-design of digital services, published by Arcep, Arcom and ADEME. It provides a grid of verifiable criteria covering the strategy, specifications, architecture, content and hosting of a digital service.
Optimisation and eco-design
Processing several requests together on the same graphics processor. The fixed cost of loading the model is spread out, which lowers energy per request, at the price of a response delay that reserves this lever for uses not constrained by latency.
GreenOps for Artificial Intelligence
Filtering a set of candidates from a document search to keep only the most relevant passages before injecting them into the context.
FinOps for AI
A pricing commitment (AWS) reserving compute capacity over a given term in exchange for a substantial discount against the on-demand rate.
Adjusting the size or level of a resource to its actual observed usage.
Public Cloud
Assigning each task to the minimal model level reaching the required quality threshold.
FinOps for AI
Management of software usage rights and contractual compliance.
A pricing commitment (AWS, and by extension other providers) covering a volume of spend rather than a specific instance, offering more flexibility than a Reserved Instance in exchange for a generally slightly lower discount.
Direct emissions from sources owned or controlled by the organisation.
GreenOps & Green IT
Indirect emissions from purchased energy consumed by the organisation.
GreenOps & Green IT
Indirect emissions from the value chain, where most of the IT footprint sits: Cloud services, SaaS, end-user devices.
GreenOps & Green IT
Emissions from purchased goods and services. This is where Cloud consumption sits: the provider's Scope 2, the electricity consumed by its data centres, becomes the customer's category 1 directly, so the customer inherits the provider's accounting method, boundary and allocation key.
GreenOps & Green IT
Use of AI tools outside the perimeter approved by the company.
Applications or services subscribed outside the purchasing channels governed by the organisation.
SaaS & Licensing
Embedding cost, security and sustainability requirements from design onwards.
Reporting of cost to consuming entities for visibility, without effective budget charging.
FinOps & Budgets IT
The unique identifier of a pricing reference at a Cloud provider, matching a precise combination of service, region and configuration. SKU proliferation is one of the complexity factors specific to Public Cloud.
A standardised indicator of emissions per software functional unit.
A set-up chaining several model calls, often with external tools, to complete a task. The number of calls per completed task becomes the steering unit, for cost as much as for footprint.
GreenOps for Artificial Intelligence
The real return across all pricing commitments, beyond the displayed discount rate.
The share of calls effectively benefiting from the reduced caching rate, an advanced indicator of architectural drift.
FinOps for AI
The share of a subscribed commitment actually consumed.
Distributing data across several storage tiers (high performance, standard, cold or archive) according to its criticality and actual access frequency, so that storage cost aligns with data value.
An architecture reserving advanced models for orchestration and mobilising lightweight models for execution agents.
FinOps for AI
The target organisation describing roles, processes and governance bodies.
The elementary billing unit of language models, corresponding to a fragment of text processed as input or produced as output.
FinOps for AI
A token sent to the model: system instructions, conversation history, tool definitions, retrieved documents. Serves as the reference rate.
FinOps for AI
A token generated by the model during its internal reasoning, billed at the output rate while remaining absent from standard dashboards.
FinOps for AI
A token produced by the model, billed at a rate generally several times higher than the input rate.
FinOps for AI
An input token corresponding to an already processed prompt prefix, charged at a fraction of the standard rate.
FinOps for AI
The discipline of analysing and governing the costs linked to token consumption and the billing mechanics specific to generative AI.
FinOps for AI
A native platform guardrail setting an automatic maximum lifetime for a resource (cluster, test environment), beyond which it is stopped or deleted, limiting consumption drift caused by forgotten resources.
Virtual processor, the billing and sizing unit for Cloud compute. Its energy draw is estimated from average power coefficients (Cloud Carbon Footprint model), around 0.7 W idle and 3.5 W at full load for an x86 vCPU: indicative values, since the gap between processor generations can exceed a factor of two at comparable performance.
GreenOps & Green IT
A shared compute unit specific to Snowflake, whose sizing and scaling behaviour are direct cost optimisation levers.
A shared compute unit specific to Databricks, whose sizing and scaling behaviour are direct cost optimisation levers.
Methodological framework provided by Cloud providers (AWS, Azure, GCP) to align business, cultural, and technical strategies during a Cloud adoption journey.
Cloud Center of Excellence & AI
Security and governance framework ensuring that only authorized users and systems access Cloud resources, based on the principle of least privilege.
Cloud Center of Excellence & AI
Security model assuming no implicit trust, requiring continuous verification of every identity and device, even within the corporate network.
Cloud Center of Excellence & AI
European regulatory framework classifying AI systems by risk level (minimal, limited, high, unacceptable) and imposing strict transparency and compliance obligations.
FinOps for AI
The ability to understand, trace, and justify an AI model's reasoning process, a prerequisite for regulatory compliance and user trust.
FinOps for AI
Systematic and unfair deviation in an AI model's outputs, typically inherited from unrepresentative training data or biased design.
FinOps for AI
Governance approach aimed at developing and deploying AI systems in an ethical, transparent, secure, and privacy-respecting manner.
FinOps for AI
The ability to infer the internal state of a complex system based on its external outputs (metrics, logs, traces). Goes beyond traditional monitoring by explaining why an issue occurs.
FinOps & Budgets IT
Methodical process, often accelerated by AIOps, aimed at identifying the fundamental origin of an incident rather than merely treating its symptoms.
FinOps & Budgets IT
Software engineering approach applied to IT operations, utilizing code and automation to manage system reliability, scalability, and performance.
FinOps & Budgets IT
Core DevOps practice automating code testing, validation, and delivery phases, thereby reducing deployment risks and accelerating Time-to-Market.
FinOps & Budgets IT
Operational framework using Git as the single source of truth for infrastructure and applications, where every change is versioned, reviewed, and automatically deployed.
FinOps & Budgets IT
A unit aggregating greenhouse gases according to their global warming potential over 100 years.
GHG Protocol framework
An average emission value per unit of activity, used to convert a physical or monetary figure into an equivalent emission.
GHG Protocol framework
The year used as a comparison point to measure an emissions trajectory, to be recalculated after a structural change.
GHG Protocol framework
The purchase of credits representing an emission reduction or removal achieved elsewhere, reported separately from gross emissions.
GHG Protocol framework
Removal of CO₂ from the atmosphere by a natural or technological mechanism, distinct from offsetting.
GHG Protocol framework
The ability to run a use case on an alternative model without rewriting integration code or fully recalibrating prompts.
Artificial Intelligence & GenAI
A design that decouples application logic from the underlying model through an abstraction layer, so workload can move from one provider to another.
Artificial Intelligence & GenAI
Modelling the profitability of a use case under several price assumptions: current rate, announced rate and a plausible intermediate rate.
Artificial Intelligence & GenAI
Technical or contractual dependency that makes switching provider, model or platform costly.
Artificial Intelligence & GenAI
A single point of passage for model calls, providing attribution through virtual keys, spend capping and switching between providers.
Artificial Intelligence & GenAI
A deliberately low rate applied at a model's launch to accelerate adoption, raised towards a target rate once the installed base is locked in.
Artificial Intelligence & GenAI
European directive 2023/1791 which requires data centres with at least 500 kW of installed IT power to report annually their energy consumption, PUE and water use to the dedicated European database.
Footprint measurement
Mechanism allowing a software licence already owned to be reused on Cloud infrastructure, instead of paying for the licence embedded in the provider rate. Its value depends on the mobility rights written into the contract and on the associated compliance tracking.
FinOps & Budgets IT
Training a smaller model to reproduce the behaviour of a larger one on a given domain. The resulting model consumes fewer resources per request, at comparable quality on the targeted perimeter only.
Artificial Intelligence & GenAI
Certificates attesting that a volume of renewable electricity was injected into the grid. They feed the market-based method in Scope 2 accounting, without changing the electricity physically consumed on site.
Energy sources and location
Condition of a territory whose water withdrawals approach or exceed the available renewable resources. This criterion qualifies the real impact of a litre consumed by a data centre depending on its location.
Energy sources and location
Applying standardised metadata to resources at creation time, so that each cost and each consumption is attached to an accountable entity. Without tagging applied at deployment, allocation becomes a manual reconstruction.
FinOps & Budgets IT
End of use of a device driven by failure, the end of software support or the renewal policy alone. The software and contractual share often dominates the hardware one, which makes it a decision lever rather than a technical inevitability.
Hardware and end-user devices
Equipment restored to working order and resold with a warranty after traceable erasure of the previous holder's data. Its manufacturing footprint already being amortised, it avoids most of the embodied carbon of a new device.
Hardware and end-user devices
Reference framework published by the FinOps Foundation, structured in principles, personas, phases (Inform, Optimize, Operate), domains and capabilities. It provides the shared vocabulary and maturity grid of a FinOps practice.
FinOps & Budgets IT
A provider's contractual commitment on a measurable service level, availability or response time, backed by penalties or credits when it is not met. Those credits are a recovery lever that often goes unclaimed.
Execution model in which the provider allocates resources on demand and bills actual usage, with no capacity reserved by the customer. Spend follows traffic, which shifts optimisation towards code design and call volume.
Isolated execution unit bundling an application and its dependencies while sharing the host kernel. Since billing applies to the underlying machine, cost allocation requires per-container consumption measurement.
Container orchestrator that schedules workloads across a pool of machines. Its bill appears at machine level rather than application level, which makes it one of the hardest perimeters to allocate and one of the richest in idle capacity.
FinOps & Budgets IT
Reuse of the computation already performed on the identical opening segment of several requests sent to a model, system instructions and shared context. The tokens concerned are billed at a reduced rate and are not recomputed.
Artificial Intelligence & GenAI
Ability of an infrastructure to scale resources with demand, upwards and downwards. It only delivers a gain if the downward move is actually applied, which requires thresholds and automated shutdown.
Compute capacity sold at a heavily reduced rate from a provider's unused resources, reclaimable at any time with short notice. It suits interruptible, replayable processing, not sensitive production workloads.
Mechanism that automatically adjusts the number of instances or the size of a service according to load indicators. Poorly configured, it durably settles the infrastructure on its peak rather than on its real need.

FAQ

Frequently asked questions

Cost reduction is a frequent outcome, and it should remain one objective among several. FinOps produces above all the ability to arbitrate: to decide knowingly between performance, delivery speed, risk and spend. A well-informed FinOps decision can lead to deliberately increasing a cost.

Both can progress in parallel. Cloud governance does strengthen the anchoring: with named owners and deployment standards, optimisations hold over time rather than being erased by subsequent deployments.

They share their first levers, removing the unused and rightsizing, then diverge on certain low-carbon trade-offs that carry a cost. Handling them within one governance framework makes those trade-offs explicit.

By reasoning in unit cost per use case rather than in a global envelope, instrumenting allocation from the first pilot, and treating product design as the primary cost lever, ahead of model selection.

AIOps uses AI to operate the information system. AI governance frames the company's AI usage, including that of AIOps.