Inference Public

Cloud foundations

Planning an OU strategy that outlasts the next reorganisation

Organisational units decide how policy reaches every account you own. A structure that copies the organisation chart has to be redesigned whenever the chart changes.

A hand rearranging plain dividers between three organised trays.

Ask which accounts need the same rules

In AWS Organizations, policies attached to an organisational unit apply to every account beneath it. The useful design question is therefore which accounts should be governed in the same way, which is a different question from who manages whom. Production workloads that handle personal data need stricter controls than a sandbox, whichever directorate owns them.

Departmental ownership, cost centres and service names still matter, and they are better recorded in account tags and naming, which can change cheaply when the organisation does.

A sensible starting set

AWS's guidance on organising multiple accounts groups its recommended OUs by purpose. Foundational OUs hold security accounts and shared infrastructure such as networking. A workloads OU holds business services, usually separated into production and non-production. A sandbox OU provides room for experiments with limited reach into production.

It also describes procedural OUs: one for exceptions to the standard policies, one for accounts in transition, one for suspended accounts and one for testing new policies before they reach production. AWS is clear that you do not need all of them at once, so start with what the estate needs and add the rest as requirements appear.

Design within the limits

AWS Organizations allows OUs to be nested five levels deep, with up to ten service control policies attached to each OU or account, and each policy limited to 10,240 characters. Those limits are generous for a sensible design. Deep nesting makes the combined effect of inherited policies hard to reason about, so a shallow structure is usually easier to explain to an auditor and to a new engineer.

Plan the procedural OUs early

A policy staging OU lets you test a new service control policy on representative accounts before it blocks a production deployment. A transitional OU gives accounts inherited from a supplier, a predecessor body or a merger somewhere to sit while they are brought up to standard. A suspended OU holds accounts that are awaiting closure, with almost everything denied. An exceptions OU makes deviations visible, and every account in it should have an owner and a review date.

Public-sector considerations

Where different data handling rules apply, such as services processing sensitive personal data, a separate OU with stricter controls can be easier to assure than exceptions scattered across the estate. Where an organisation provides shared services to other public bodies, decide whether those bodies' workloads belong in your structure at all, and record who is accountable for each one.

Working document: an OU design record

A short record for each OU explains its purpose to the people who will maintain it long after the initial design.

Illustrative template
FieldWhat to record
OU and pathName and position in the hierarchy.
PurposeWhy these accounts are governed together.
Entry criteriaWhat an account must satisfy to be placed here, and who approves it.
PoliciesService control and resource control policies attached, with the reason for each.
ControlsDetective and proactive controls enabled at this level.
OwnerThe person accountable for the OU's policies and membership.
Exit routeWhere accounts go when their purpose changes or they are closed.
Review dateWhen the design and its exceptions will next be reviewed.

For a worked example, see machinery of government changes and AWS accounts, or discuss your AWS foundations.

Further reading

Organizing your AWS environment using multiple accounts: recommended OUs and accounts

AWS Organizations: quotas

Related articles

Where Control Tower fits in a cloud operating model

A landing zone needs owners as well as accounts.

All insights

Start a conversation

Let’s talk about what you need to build.

Bring the service, the constraint or the engineering question. We’ll work through a sensible starting point.

Discuss a project