TECHNOLOGY & CERTIFICATION EDITORIAL

Microsoft AZ-104: Administrator Skills and Exam Preparation

The Azure Administrator exam rewards people who can explain what a configuration change will do before they make it. A candidate may know how to create a virtual machine yet struggle to determine why an apparently correct role assignment does not grant blob access, or why a private endpoint fails even with valid credentials. The skill being measured is operational judgment across interconnected Azure services.

A useful preparation plan for the Microsoft AZ-104 exam begins with decisions an administrator has to defend. An identity may have permission to create a storage account but not read its blobs; a private endpoint may exist while a client still resolves the public address; a VM backup may be successful without proving that the application can be restored. Microsoft’s April 17, 2026 objectives group those responsibilities into identities and governance, storage, compute, networking, and monitoring with recovery. Practice each domain as a change, a likely failure and a verification step. That produces stronger evidence of readiness than memorizing which portal blade contains a setting.

What the certification proves and what it does not

AZ-104 is an intermediate Azure administration exam, not a substitute for professional experience operating production systems. It validates that a candidate understands resource configuration and can reason about permissions, networking, workload capacity and recovery. Passing does not mean the person has designed every possible Azure architecture or can approve major security changes without organizational review.

Microsoft expects familiarity with operating systems, networking, servers and virtualization, together with hands-on use of Azure portal, Azure CLI, PowerShell, ARM or Bicep and Microsoft Entra ID. If these terms are completely unfamiliar, a fundamentals path can help establish vocabulary, but repeated portal clicks without understanding scopes and effects are insufficient for the administrator role.

The most useful evidence of preparation is not the number of videos watched. It is a set of scenarios the candidate can explain and reproduce: correctly scoped access, a recoverable storage account, a functioning private network, a deployment rollback and a tested backup restore. Those are the experiences interviewers and operational teams can meaningfully discuss.

Identity and governance: 20–25 percent

The identity domain covers creating and maintaining Microsoft Entra users and groups, managing licenses and guests and configuring self-service password reset. It also covers Azure role assignments and resource governance, including policy, tags, locks, subscriptions, management groups and cost controls. The challenge is deciding which control solves the scenario, because an authentication problem, an RBAC denial and a policy refusal can look similar to an end user.

In a test Azure subscription, create a user, assign a narrow role at a resource-group boundary, then inspect which inherited permissions remain. The differences between Entra users and groups and Azure RBAC scope behavior become clear when that user can view a resource but cannot read its storage data. Record the expected access outcome, use a harmless test account, and reproduce a policy denial before changing governance settings.

Resource placement is a financial and governance decision, not just an administrative naming choice. Subscription budget alerts can warn administrators about increasing spend, but cannot themselves stop charges or restore deleted data. In a lab, trigger a budget threshold, attempt a blocked deletion on a disposable resource, and document which control changed behavior and which merely informed an operator.

Storage: 15–20 percent

Storage objectives include account configuration, network restrictions, SAS tokens, shared keys, identity-based Azure Files access, redundancy, replication, Blob and file-share features and utilities such as AzCopy and Storage Explorer. Many mistakes arise from assuming a storage account’s network path and its authorization mechanism are the same thing. They are not.

Configure a storage account in an isolated lab and test a client that succeeds and one that fails. Compare a short-lived shared access signature with an Entra-authorized data role and an account key, then test Blob and Files lifecycle behavior after a soft-delete or versioning operation. When the client cannot read the object, distinguish a data-plane permission error, network restriction, deleted version or retention policy before changing settings.

Local, zone and geographic storage redundancy reduce different infrastructure-failure risks. Geographic redundancy and transfers may involve asynchronous replication and are not a substitute for recovering from accidental deletion or logical corruption. Document which failure your chosen redundancy mode covers, then test restoration from a known recoverable object state rather than treating a healthy replication indicator as proof of application recovery.

Compute: 20–25 percent

Azure compute administration is broader than creating VMs. The guide includes ARM and Bicep changes, virtual machine sizes and disks, encryption at host, availability sets and zones, scale sets, Azure Container Registry, Container Instances, Container Apps and Azure App Service plans, TLS, networking, backup and deployment slots.

Use an infrastructure-as-code exercise to deploy a constrained environment, run what-if and review the result. Treat Bicep and ARM deployments as a reproducibility exercise and VM availability and scale sets as a failure-tolerance exercise; a deployment template alone proves neither availability nor recovery. Add a disk-resize or encryption-at-host scenario only after checking supported prerequisites.

Then contrast two managed application paths: ACR, ACI and Container Apps for image lifecycle, revisions and scaling, and App Service deployment slots for a web application release. These services may all host code, but their deployment, traffic and identity models differ. A lab should make those differences observable.

Networking: 15–20 percent

Virtual networking objectives include address spaces, subnets, peering, public IPs, user-defined routes, troubleshooting, NSGs, application security groups, Bastion, service and private endpoints, Azure DNS and load balancing. Learn to trace the request path: name resolution, route, security decision, listener, backend health and authorization.

Start with VNets, peering and routes to understand how address prefixes and next hops govern reachability. Then use NSGs, Bastion and private endpoints to examine access protection. A private endpoint is not a replacement for identity checks and often depends on a correct DNS path.

Azure DNS and load balancing answer different questions: DNS resolves the name to a reachable address, while a load balancer distributes traffic among backends that satisfy its configuration and health checks. A correctly configured load-balancer frontend does not guarantee that a health probe finds a ready backend. Include a lab that deliberately breaks resolution and another that breaks a probe, and explain why the remediation differs.

Monitoring and recovery: 10–15 percent

Azure Monitor metrics, logs, alerts, Insights and Network Watcher provide different evidence. A metric may show rising latency; logs may identify failed requests; Connection Monitor and network diagnostics help determine whether the underlying path is reachable. An effective administrator collects enough evidence to choose the right change instead of scaling every service that appears slow.

Azure Monitor and Network Watcher supplies the diagnostic framework. Reproduce one network failure, inspect effective routes and security rules, and compare the event with resource health signals. Use a realistic alert threshold and confirm who receives the notification. An alert that nobody can act upon is an incomplete operational control.

In Backup and Site Recovery scenarios, distinguish recovery-point retention, the applicable vault and policy, a successful restore, and a supported failover path. Test a point-in-time restore and distinguish it from high availability within one region or replication of a running workload. Recovery is a business outcome: a backed-up service is not proven recoverable until a suitable restoration path has been exercised.

A practical eight-week preparation sequence

An eight-week plan can be useful if it is treated as an adaptable example rather than an official guarantee. Begin with an environment check and core identity work, then spend focused time on governance and storage. Use the middle weeks for compute and networking labs, and reserve the final portion for monitoring, restore exercises and scenario-based review. A candidate with a strong networking background may need more time for identity or App Service instead.

Turn each scheduled exercise into a falsifiable result: deploy an identity assignment expected to permit a management action but reject data-plane blob access; configure a private endpoint and deliberately misconfigure DNS; restore a virtual machine and verify application state rather than simply observing a green job. Save the original configuration and your expected result before testing. When you can explain the exact access or forwarding boundary that changed, the lab provides evidence for readiness instead of time spent in the portal.

For each lab, record the intended state, steps, observed result and recovery action. Test both successful configurations and deliberately denied requests. The second kind of test builds the habit of locating the correct layer. Take concise notes about role scope, address prefixes, service availability restrictions and data recovery expectations, not long summaries of every feature page.

Use current Microsoft Learn practice resources and the official study guide as the reference for what the exam measures. Avoid treating third-party question collections as evidence of the actual exam bank, and do not assume memorizing a question answer proves that a configuration choice is understood. When uncertain, recreate a harmless example in your own account and inspect the resource response.

Readiness means being able to explain consequences

A candidate who can define LRS but cannot distinguish it from backup needs more work. The same applies to someone who can create an NSG but cannot describe effective rules, or who can swap an App Service slot without knowing which settings are sticky. Build brief scenario prompts that force you to justify a choice and name at least one reason an alternative would fail.

Microsoft’s published weights help allocate preparation effort but should not be converted into a promise about the exact number of questions in a particular session. Role-based exams can include interactive or case-style situations and are updated as services evolve. Verify current scheduling, language, time and renewal rules on the official certification page rather than preserving an old promotional figure in the article.

The objective of preparing for AZ-104 should be twofold: pass the assessment and develop an administrator’s habit of tracing dependencies. The habits built through scoped changes, meaningful monitoring and practiced recovery are useful after the exam, when configuration decisions affect real workloads instead of isolated lab resources.

Use exam objectives as operational questions

Instead of turning the official outline into a checklist of service names, translate each objective into an administrative situation. For “configure name resolution,” ask what the client resolves from inside and outside a linked virtual network. For “interpret role assignments,” ask how a user can inherit access without a direct assignment. For “perform backup and restore,” ask what business state must be recovered and how long the return to service is allowed to take.

This approach keeps preparation practical and prevents a candidate from confusing feature recognition with implementation skill. The administrator who can describe both the intended change and the verification step is much closer to the responsibility the credential represents.

Back to Insights
Explore what matters. Knowledge that goes beyond the exam.
Explore ExamTopics