Australian Government Architecture
Search

Type

Design

Reference

Content status

Current

Compliance

Non-mandatory

Reuse category

Guidance

Guide to managing cloud and usage costs, including AI costs

View more Cloud policy guidance.

On this page

About this guidance

This guidance helps agencies meet requirement 4 of the Whole-of-government cloud computing policy: Entities must actively manage and optimise cloud computing costs.

It provides practical steps agencies can take to make cloud costs visible, accountable and manageable before, during and after migration.

The guidance is based on a FinOps approach to cloud cost management. FinOps brings finance, technology and business owners together so agencies can understand cloud demand, forecast expenditure, manage variances and optimise services over time.

Agencies can apply this approach through a recognised FinOps framework, business technology management practices, provider-native tools, or existing governance, finance, ICT and product management processes.

Who this guidance is for

This guidance is for agency staff involved in planning, funding, approving, delivering, operating or reviewing the costs of cloud services. Different roles will use the guidance to make sure cloud expenditure is visible, accountable, forecast and optimised throughout the service lifecycle. 

  • Accountable executives — set clear expectations for cloud cost accountability, approve budgets and cost models, assign accountable owners and review whether cloud expenditure remains aligned with agency outcomes, value and risk appetite. 
  • Delivery and program teams — build cost planning into cloud migration, procurement, service design and implementation. This includes documenting assumptions, forecasting migration and operating costs, applying tagging and cost allocation standards, and escalating material cost trade-offs. 
  • Operational and specialist teams — manage cloud costs day to day by monitoring consumption, maintaining dashboards and alerts, reviewing access and configurations, right-sizing services and identifying opportunities to optimise cost, performance and value over time. 

When to use this guidance

Use this guidance when planning a new cloud service, expanding an existing cloud service, migrating from legacy arrangements, or reviewing whether cloud services remain necessary, efficient and aligned to business outcomes.  

Agencies should ensure consultation with relevant finance and budget areas are done when considering the planning, funding, approving, delivering, operating or reviewing the costs of cloud services. 

Planning for cloud costs

Cloud computing can increase agency capability, but it needs careful planning to avoid unexpected costs. Factors to consider include: 

  • a shift in IT costs from capital expenditure, such as hardware ownership, to operational expenditure 
  • one-off migration costs 
  • usage-based costs that can go undetected and vary significantly if not managed appropriately 
  • a wide range of cost models, options and billing arrangements that vary by vendor. 

Early planning helps set expectations, test assumptions and make informed choices about architecture, service design and investment. 

Agencies should separate the minimum cost controls needed before adopting cloud services from the optimisation practices that continue after adoption. This helps accountable officials understand which controls need to be in place before migration and which activities should be managed as part of ongoing service operations. 

The table below sets out the minimum controls agencies should establish before adopting or expanding cloud services, and the evidence they should maintain to demonstrate these controls are in place. 

Minimum cost control  What agencies should do  Evidence agencies should maintain 
Accountable owner  Assign clear roles to finance, architecture, ICT, business and product teams so cloud demand and costs are identified, understood and owned.  Named cost owner, roles and responsibilities, escalation pathway and agreed reporting audience. 
Budget  Establish clear budgets for migration, running and exit costs, and assign responsibility for budget and cost management in line with the agency’s budgeting process.  Approved budget, budget owner, funding assumptions and distinction between migration, operating and exit costs. 
Cost model  Develop a clear architectural and cost model for which services should be cloud-based, including storage, data transfer, compute, database, API services, managed services and software licensing costs.  Cost model, vendor calculator outputs, assumptions, options analysis and documented cost drivers. 
Tagging and cost allocation  Break down costs by service, application, business unit or agency function so expenditure can be forecast, reported and attributed to the right owner.  Tagging standard, cost allocation model, service owner mapping and reporting structure. 
Access control  Ensure access control and administration are appropriately managed and users have access only to the services they need to perform their role.  Access control model, administrator roles, approval workflow and periodic access review records. 
Reporting  Establish regular reporting that gives finance, technology and business owners visibility of forecast and actual expenditure before migration begins.  Reporting cadence, dashboard or report template, accountable recipients and variance escalation process. 

Governing cloud costs

Clear governance arrangements support accountability for cloud expenditure and help agencies manage costs and identify opportunities to optimise resources. 

The following table sets out core FinOps controls agencies should establish to make cloud expenditure visible, accountable and manageable.  

After adoption, agencies should use these arrangements to govern demand, allocation, forecasting and optimisation: 

Ongoing governance area  What agencies should maintain 
Demand and allocation  Policies for provisioning, showback or chargeback, and allocating expenditure to the relevant business owner. 
Forecasting and variance management  Regular comparison of forecast and actual expenditure, with thresholds and escalation for material variances. 
Optimisation decisions  A prioritised record of optimisation actions, benefits, trade-offs, owners and completion dates. 
Governance review  Periodic review of whether cost controls, reporting and accountability remain effective as services change. 

Monitoring and optimisation

Ongoing monitoring and optimisation help agencies maintain visibility of cloud expenditure and respond early to unexpected changes in use. Regular oversight supports forecasting, highlights unexpected consumption, enables automated cost controls and improves efficiency. 

  • maintain cost visibility across business units, agency functions, applications and services before and after migration 
  • compare forecast and actual costs, monitor utilisation and track cost trends to identify variances, underused resources and emerging cost pressures 
  • use dashboards, alerts, automated controls and regular configuration reviews to detect unexpected increases early and prevent avoidable overspend 
  • review services, access, test environments and scaling settings regularly so resources are only active and configured when needed 
  • optimise continuously by tracking high-consumption workloads, data platform costs, expensive queries and the impact of optimisation on performance, user experience and outcomes. 

AI FinOps and token economics

Cloud platforms can provide greater access to AI services, but AI usage costs are often billed separately from cloud compute, storage and data transfer.  

Agencies should understand how each cost model works before adopting AI-enabled services. 

Apply the FinOps approach described above to AI services, while accounting for their distinct consumption metrics, lifecycle costs and pricing volatility. 

Use of AI on cloud introduces additional challenges because pricing models are more varied and volatile consumption metrics apply. 

Many AI services use consumption-based billing based on tokens. Tokens are units of text or data that an AI model processes when receiving an input, generating an output or handling supporting context.  

Token use is a form of consumption, similar to utility use. It needs to be visible, controllable and accountable so agencies can manage cost, value and demand over time. Token-based usage and pricing is generally applicable when accessing models provided by vendor AI Services. 

Many cloud cost management principles also apply to AI usage, but where appropriate, agencies should treat AI charges as a distinct cost category to better manage overall costs. 

Agencies should make AI consumption visible, budgeted, accountable and controlled before scaling AI services across the enterprise. 

The following table provides a checklist for AI cost governance controls. 

Checklist item  What agencies should establish  Evidence agencies should maintain 
AI service ownership  A named accountable owner for each material AI service or AI-enabled workflow.  Service owner, business owner, cost owner and escalation pathway. 
AI consumption model understood   A documented view of how each AI service is charged, including tokens, model calls, context windows, agent execution and downstream service calls where relevant. The consumption model and associated budget or forecast must account for the full AI lifecycle, including experimentation, training, AI model lifecycle, and related assurance activities, not only production workloads. Costs across pre-production phases can be significant and should be identified and attributed and from the outset.  Charging model summary and assumptions used in forecasting, covering both production and pre-production activities. Evidence should demonstrate that lifecycle costs, including experimentation, evaluation, training, and assurance, have been considered and where applicable attributed to a cost centre or project. 
Budget and forecast  A separate forecast for AI consumption costs before enterprise deployment.  Approved budget, forecast consumption range and assumptions about user numbers, transaction volumes and model use. 
Usage visibility  Regular monitoring of actual AI usage and expenditure. Apply resource tagging to all AI services, attributing expenditure to the business function or cost centre it supports.  Consumption reports, dashboards, alerts and variance analysis. 
Unit cost and value  A way to understand the cost of the AI-enabled service against a business outcome, based on total cost rather than model usage alone, including compute, storage, retrieval and vector databases, knowledge base stores, orchestration, monitoring and logging, evaluation, and downstream cloud services. These components are typically decoupled across multiple services and providers.  Cost per transaction, user, request, workflow or other relevant outcome, calculated across all contributing components and services rather than model usage alone. 
Guardrails and limits  Controls to prevent uncontrolled or unintended consumption.  Usage thresholds, rate limits, quotas, approval gates, model access controls and escalation triggers. 
Benefits review  Periodic assessment of whether realised benefits justify ongoing AI consumption costs.  Benefits review, cost trend, usage trend and decision to continue, optimise, limit or retire the service. 

Related links

More cloud policy guidance

Was this information helpful?

Do not include any personal information. We are unable to respond to comments or feedback. If you would like a response, please email, or phone us. Our details are on the AGA contact page www.architecture.digital.gov.au/contact-us.