Skip to content
← All articles
6 min readCloud modernization

Treating a cloud landing zone as a product

A landing zone is not a finished platform or a one-click template. It is the set of decisions that determines how predictably teams can build, operate, and change workloads in Azure.

A landing zone is a set of decisions

It is tempting to treat an Azure landing zone as an installation task: pick a reference architecture, apply the templates, hand over the diagram. That produces resources, but not necessarily a platform. The useful question is not whether the landing zone includes management groups and policy. It is whether a team can answer basic questions without a meeting. Where does a new workload go? Who approves a subscription? Which network path does it use? What happens when a control blocks a deployment?

Those answers are decisions about boundaries, ownership, and change. Azure supplies the building blocks: management groups, subscriptions, policies, role assignments, virtual networks, private DNS zones. A landing zone is the arrangement of those blocks that makes the next hundred decisions easier rather than harder. The arrangement should be explicit, documented, and testable.

Infrastructure as code is what makes that arrangement reviewable. When the landing zone lives in Terraform or Bicep, a change to a policy assignment or a DNS link becomes a plan with a reviewer, not a portal action with a screenshot. The code is not the platform, but it is the record of how the platform is assembled. Treat the pipeline that applies it as part of the landing zone, with the same ownership and approval rules.

Start with the people who use the platform

A landing zone has several users: product teams who need somewhere to deploy, operations who support what runs there, security and compliance who need evidence that controls work, and finance who need costs attributed. Each sees a different surface. If the platform is designed only for the platform team, the first product team becomes the integration test.

A practical exercise is to walk through three journeys: a team onboarding its first workload, an engineer requesting an exception, and an auditor asking who changed a control. If any journey depends on knowledge held by one person, the platform has a bus-factor problem before it has a scale problem. Write the journeys down and make them part of the platform’s acceptance criteria, not an afterthought.

Subscriptions and environments: boundaries that mean something

Subscriptions are often treated as environments: development, test, production. Sometimes that is right. But a subscription is also a scale unit, a policy and access boundary, and a billing boundary. When those roles pull in different directions, naming will hide the conflict rather than resolve it.

Decide what a subscription represents before deciding how many you need. A common pattern is one subscription per workload or product group, with production and non-production separated by policy and access rather than by a proliferation of subscriptions. Whatever the pattern, the rule should hold: a subscription boundary should correspond to a real ownership, policy, or cost boundary. If it does not, it is a folder with a bill.

Identity and network are the hard-to-change decisions

Compute can be rebuilt. Identity and network topology are expensive to reverse. Entra ID tenant structure, privileged access, workload identities, hub-and-spoke versus Virtual WAN, address space, private endpoints, and private DNS zones all shape what is possible later. These deserve slower decisions and better records.

Private DNS is a good example of a detail that behaves like architecture. A private endpoint with incorrect resolution fails in ways that look like application errors. The landing zone should define where private DNS zones live, how they are linked, how on-premises resolution reaches them, and who owns record changes. The same is true for identity. Decide where human and workload identities come from, how privileged roles are activated, and which identities can change platform controls. If the answer is the platform team’s standing admin account, the design has a concentration risk.

Policy should explain itself

Policy is where a landing zone becomes visible to engineers. A denied deployment is a conversation. If the denial names a policy but not the reason, the supported alternative, or the owner, the engineer learns to work around the platform rather than with it. Azure Policy effects such as Audit, Deny, and DeployIfNotExists are not just enforcement settings. They are part of the platform’s interface.

Name policies for the behavior they protect, scope them where they are true, and test them against representative deployments before enforcing. Prefer a small number of meaningful initiatives over a large collection of overlapping definitions. Where a control is not ready to enforce, audit first and publish the path to enforcement. The goal is not a policy count. It is a control set engineers can predict.

Ownership is the part templates cannot supply

A reference architecture can describe a platform team, but it cannot create one. Someone must own the management group hierarchy, the policy definitions, the network hubs, the private DNS zones, the subscription vending process, and the exception queue. Those responsibilities need names, not team names. They also need a review cadence, because a landing zone that is correct in March may be wrong in September.

Keep a decision record for each significant choice: the context, the options considered, the decision, and the condition that would reopen it. This is not documentation for its own sake. It is how a new engineer, an auditor, or a future you can tell whether a control is intentional or accidental.

Product thinking also applies to the platform’s roadmap. A landing zone needs an intake path for new requirements, a way to prioritize them against operational work, and a release note when a guardrail changes. Without that, the platform team becomes a ticket queue and the platform stops evolving. A short monthly review of requests, exceptions, and incidents is often enough to keep it honest.

What the landing zone should make easier

A useful test is to list the things the platform should make easier and check whether it does. Onboarding a workload should follow a known path. Finding the owner of a control should take minutes. Requesting and receiving a legitimate exception should be possible without bypassing the platform. Producing evidence for an audit should not require reconstructing history from chat.

If the current landing zone makes those things harder, more templates will not fix it. Start with the decision that causes the most repeated friction, make it explicit, and improve the platform around it. That is slower than a one-click deployment and considerably faster than rebuilding the platform twice.

Measure the platform by the work it removes, not the resources it creates. Time to onboard a workload, time to resolve an exception, the number of manual steps in a standard change, and the proportion of controls with a named owner are all observable without inventing a business case. Review them with the teams who use the platform and publish the trend. A landing zone that is improving in those terms is doing its job, even if the diagram has not changed.

Related service

Cloud modernization

Next step

If this is the decision in front of you, describe the constraint. I will tell you whether the work fits.

Start a project inquiry