Organisations are actively pursuing data and AI initiatives, but for many there’s a disconnect between their ambitions and their cloud readiness. We recently asked CIOs if their cloud estate was ready for AI, and now we’re providing practical checks across the same visibility, data, architecture, governance and economic foundations.
The success of your data and AI initiatives will be influenced by the cloud infrastructure beneath them. This 20-point checklist helps identify potential weaknesses in your Azure environment, and outlines steps you can take to address them.
Visibility: know what you have and how it’s being used
Before you can optimise, you need to understand what you’ve got.
1. Do significant Azure resources and workloads have identifiable owners?
Ownership creates accountability. Define an owner for every significant workload and make ownership part of your tagging standard. There should be clear accountability for why resources exist, whether they are still required and whether their performance and costs remain appropriate.
2. Can you identify your biggest sources of Azure consumption?
Use Microsoft Cost Management to break down expenditure and identify which services, workloads and business activities are driving consumption – and monitor how these change over time.
3. Are resources consistently tagged and organised?
Tagging makes it easier to understand the purpose, ownership and cost of Azure resources. Define the information you need – such as workload, owner, environment and cost centre – and use Azure Policy to apply these requirements consistently.
4. Can you identify under-utilised, redundant or legacy resources?
Azure Advisor and Azure Monitor can help identify idle or consistently under-utilised resources. Establish whether they can be resized, consolidated or retired, and don’t overlook forgotten development and test environments.
Data: make the right data usable
AI needs more than lots of data. It needs the right data to be discoverable, trustworthy and efficiently accessible.
5. Do you know which data your priority AI use cases need?
Don’t expect to make all your organisational data ‘AI-ready’. Start with your priority use cases and identify the data required, where it resides, who owns it and whether it is accessible and fit for purpose.
6. Can workloads access that data efficiently?
Understand how data reaches your applications, analytics platforms and AI workloads. Look for unnecessary movement, duplication or transfers between environments that could add latency, complexity and cost.
7. Are you storing more data than you need – or storing it inefficiently?
Identify unnecessary copies and data being retained beyond its useful life. Review how frequently retained data is accessed. Move infrequently used data to lower-cost storage, deleting it when it is no longer required – subject to regulatory, legal and business retention requirements.
8. Is your data trustworthy and appropriately governed?
Ensure that data is accurate, current and sufficiently well understood for its intended use. Identify sensitive and business-critical data, establish who owns it and ensure access is appropriate.
Architecture: make sure infrastructure can adapt
An environment designed around yesterday’s applications won’t automatically be right for tomorrow’s data and AI workloads.
9. Is compute matched to actual workload requirements?
Use utilisation and performance data to identify over- or under-provisioned resources and right-size them accordingly. For AI workloads, check that expensive specialised compute is genuinely required and appropriately sized rather than simply becoming the default.
10. Can workloads scale up – and back down – with demand?
Avoid permanently provisioning capacity for occasional peaks. Where workloads permit, use autoscaling or scheduled scaling to increase capacity when required and reduce it – and the associated cost – when demand falls.
11. Are data and compute appropriately located relative to each other?
Review where data is stored and processing takes place. Repeatedly moving large datasets between locations can increase latency, complexity and cost, so consider whether processing can take place closer to the data it uses.
12. Is there a defined route from experimentation into production?
Be clear from the outset how successful AI experiments will progress into production. Define the architecture, security, governance, resilience and operational requirements they must meet before they scale.
Governance and security: put the right controls in place
Effective AI governance isn’t simply about restricting what people can do. It should enable them to experiment and innovate safely, within clear boundaries around data, access, security and risk.
13. Are identity and access permissions appropriate and regularly reviewed?
Apply least-privilege access to users, applications and AI workloads, ensuring each has access only to the resources and data it needs. Review permissions regularly and remove access that is no longer required – particularly temporary project access that has quietly become permanent.
14. Are governance policies applied consistently?
Use Azure Policy and your landing-zone architecture to apply common requirements for resource configuration, permitted locations, tagging and security. Automating these controls gives teams defined boundaries and avoids recreating governance requirements project by project.
15. Can you detect and respond to activity outside your governance controls?
Monitor Azure Policy compliance to identify resources that breach governance requirements, and use Microsoft Defender for Cloud to flag security vulnerabilities, misconfigurations and suspicious activity. Configure alerts and assign responsibility for resolving issues.
16. Does your Azure landing-zone architecture still reflect your requirements?
Review your landing-zone architecture at least annually, and whenever there is a significant change in your Azure estate or AI ambitions. Check that its governance, security, identity and networking arrangements support your current and planned AI workloads, and update them when requirements change.
Economics and operations: optimise continuously
Cloud optimisation isn’t simply about reducing spend. The objective is to achieve the required performance, security, resilience and business value at an appropriate cost.
17. Are you paying for resources when they aren’t being used?
Look for resources that continue consuming budget without delivering value. Where appropriate, schedule development and test environments to shut down when not required and reduce or suspend expensive compute during periods of inactivity.
18. Do you understand what an AI workload will cost in production?
A successful proof of concept establishes that something works – not whether it will be worthwhile at scale. Before moving into production, model costs at realistic usage levels, including compute or model consumption, storage, data processing, retrieval, monitoring and networking.
19. Can you relate the cost of an AI workload to the value it delivers?
Don’t measure AI workloads solely by their Azure consumption. Establish meaningful unit measures – such as cost per transaction, document processed or user served – and track how they change as workloads scale. Where possible, relate these costs to measurable business outcomes.
20. Is Azure optimisation an ongoing discipline?
If not, it needs to be. The alternative is waiting for an issue and continuously review cost, utilisation, performance and architecture. Use tools such as Azure Advisor and Cost Management to identify opportunities, but assess and prioritise them holistically so that improvements in one area don’t create unintended consequences elsewhere.
Is your Azure environment data and AI ready?
The objective is an Azure environment in which data is accessible and trusted, infrastructure can adapt to changing demands, governance enables safe innovation, and cloud consumption remains aligned with the value being delivered. Just as importantly, it needs to keep evolving as your workloads, data and AI ambitions change.
If the checklist has exposed gaps in your foundations, the next step is to understand which matter most and where action will make the biggest difference.
Talk to one of our Azure experts about your current environment and your data and AI plans. We can help you identify the priority steps needed to create stronger foundations for what comes next.
