AWS Control Tower and landing zones: build it, buy it or start lighter
Understand what AWS Control Tower creates in your organisation, how its controls and Account Factory work, what it costs, and how to decide between Control Tower, a lite landing zone built as code, or a mix of both.
Assumes: .NET 10 LTS · AWS SDK for .NET v4 (AWSSDK.ControlTower 4.1.x, AWSSDK.Organizations 4.1.x) · AWS CLI v2 · Terraform AWS provider v6 · Region us-east-1. Verified against official sources on 2026-11-18.
Learning objectives
By the end of this lesson you can:
- Describe what a landing zone is and exactly what AWS Control Tower creates: OUs, shared accounts, logging, controls, identity set-up and an account vending mechanism.
- Explain the three kinds of controls (preventive, detective, proactive) and the three guidance levels, and say which AWS service enforces each.
- Compare three paths: Control Tower, a "lite" landing zone built with your own IaC, and a hybrid, using a written decision matrix.
- Estimate honestly what Control Tower costs, knowing that it has no charge itself but turns on services that do.
- List the first 30 days of a landing-zone rollout, including the things Control Tower does not do for you.
- Run a small .NET program that reports failed or drifted controls per OU.
This is a design lesson (cost label: free). You do not enable Control Tower in the sandbox: it turns on AWS Config and other services across accounts and Regions. You build the lite version in the module lab and design the rest on paper.
Prerequisites
- Why a bank uses many AWS accounts and AWS Organizations structure.
- Landing-zone automation overview: the module 10 lesson on Account Factory for Terraform, Account Factory Customization and Customizations for Control Tower. This lesson does not repeat those automation shapes; it concentrates on the decision.
- Boundaries, SCPs, RCPs and resource policies.
- Tools: none required. The optional C# program needs a landing zone, so it is for reading unless you have one.
Why this matters in banking
By your fifteenth account, someone forgot to turn on CloudTrail in one, left a bucket without encryption in another, and created a third with the root e-mail on a personal address. A landing zone is the answer: a prepared environment where every new account starts from the same secure baseline automatically, in the right OU, with the logs flowing to the right place.
You have two honest ways to get one, and both are defensible:
- Buy the pattern. Use AWS Control Tower, which builds and governs a landing zone for you following AWS's prescriptive guidance.
- Build the pattern. Write the OUs, SCPs, central logging, Config aggregator and baseline stacks yourself in CDK or Terraform.
At a bank the question is not "which is better" but "which can I explain to the regulator and operate with the team I have". A hand-built landing zone is fully understood but costs engineering time forever. A managed one is faster to stand up but has opinions, drift rules and a pricing surface that you must understand.
The idea in plain language
A landing zone is a well-organised multi-account environment: an OU structure, a small set of shared accounts (for logs and for security tooling), a baseline of guardrails, and a repeatable way to create new accounts that arrive already compliant with the baseline.
AWS Control Tower is the AWS service that sets up and governs a landing zone for you. It does not invent new technology. It orchestrates AWS Organizations, AWS Service Catalog, AWS IAM Identity Center and other services, and keeps checking that the result has not drifted from what it set up. AWS describes building a landing zone in under an hour.
Analogy: a business park. Control Tower is the park's management company. Roads, security cameras and fire doors are already installed (shared accounts and baseline controls). House rules apply to every unit (controls). The leasing office (Account Factory) hands each new tenant a unit built to the standard fit-out. If a tenant knocks through a fire wall, the management company's inspector notices (drift detection). A "lite" landing zone is you being your own management company: more control, more chores.
What Control Tower creates
| Piece | What you get |
|---|---|
| OUs | A Security OU (required) and, optionally, a Sandbox OU, plus any others you register |
| Shared accounts | A Log Archive account (central, immutable log storage) and an Audit account (read-only cross-account access for security and audit; the Config aggregator's home) |
| Central logging | An organisation-level CloudTrail trail delivering to the Log Archive account, and AWS Config recording in governed accounts and Regions |
| Identity | IAM Identity Center set up with starter groups and permission sets ⚠ VERIFY current defaults |
| Controls | A set of mandatory controls that are always on, plus strongly recommended and elective ones you choose |
| Account Factory | A Service Catalog based template for creating or enrolling accounts with a standard baseline |
| Dashboard and drift detection | A view of accounts, controls and non-compliant resources, plus alerts when the landing zone diverges |
Verify the exact list against the Control Tower user guide; AWS adds options to the set-up wizard regularly (for example around logging and KMS configuration) ⚠ VERIFY.
What problem the AWS service solves
| Need | Control Tower feature |
|---|---|
| A secure multi-account start without writing everything | Landing zone set-up |
| Consistent rules across many accounts | Controls (formerly "guardrails") |
| New accounts that are safe by default | Account Factory |
| A view of compliance across accounts | Dashboard |
| Knowing when someone changed what CT manages | Drift detection |
| Updating the baseline over time | Landing zone updates |
The three kinds of control
| Kind | Enforced by | When it acts | Example idea |
|---|---|---|---|
| Preventive | SCPs (and RCPs) | Blocks the API call | Disallow deleting the log archive bucket's policy |
| Detective | AWS Config rules | Reports after the fact | Detect unencrypted EBS volumes |
| Proactive | CloudFormation Hooks | Blocks a non-compliant resource before CloudFormation creates it | Reject a template that would create a public database |
And three guidance levels: mandatory (always on, cannot be turned off), strongly recommended and elective. Controls are also mapped to compliance frameworks in the catalogue (for example NIST and PCI DSS families) ⚠ VERIFY, but mapping is not compliance: it tells you which controls are relevant to a framework, not that you meet it.
Honest limitations
- Home Region. You choose a home Region at set-up. It cannot be changed casually afterwards ⚠ VERIFY current guidance, so decide with data-residency in mind (for a UK bank, usually
eu-west-2London). Control Tower then governs the Regions you select; other Regions can be denied with a control. - It owns some resources. Control Tower manages parts of Organizations, Config and IAM. Edit them by hand and you create drift, and landing zone updates can fail until you repair it.
- It has opinions. The OU and account shape is a starting point. If your bank needs a different shape, you spend time working around it.
- Overlap of customisation options. Account Factory Customization, Account Factory for Terraform and Customizations for Control Tower overlap. Choose one primary path per concern.
- It is a baseline, not an architecture. It does not design your network, your workload accounts or your data protection. It does not produce compliance evidence on its own.
- Cost surface. Control Tower itself has no additional charge. The services it sets up and the controls it enables do (below).
What it costs
The Control Tower pricing page states that there is no additional charge for Control Tower. Costs come from the services it uses: AWS Service Catalog, CloudTrail, AWS Config, CloudWatch, SNS, S3 and VPC (for the default Account Factory VPC). AWS Organizations and IAM Identity Center are free. Detective controls show up as AWS Config rule evaluations on your bill. Verify details on the pricing page before enabling ⚠ VERIFY.
The practical risk is not the headline price. It is that every governed account and Region gets a Config recorder and rules. Multiply by account count, Region count and number of enabled detective controls, and the per-account base cost becomes the line item. Estimate it before you commit:
- Count accounts × governed Regions.
- Estimate configuration items recorded per account per month (a quiet account records few, a busy data platform many).
- Count the detective controls you plan to enable (each evaluates per resource change).
- Multiply by the current Config prices ⚠ VERIFY.
Microsoft / Azure comparison
| You know (Microsoft/Azure) | AWS equivalent | What's different |
|---|---|---|
| Azure Landing Zones (Cloud Adoption Framework) with management groups, policies, platform subscriptions | Control Tower landing zone | Control Tower is a managed service with a console; ALZ is mostly reference architecture plus Bicep/Terraform you own |
| Azure Policy initiatives with deny/audit/deployIfNotExists | Control Tower controls (preventive, detective, proactive) | AWS splits enforcement across three technologies (SCP, Config, Hooks) rather than one policy engine |
| Subscription vending with Terraform/Bicep modules | Account Factory, AFT, AFC | AFT is the closest to "subscription vending in Terraform" |
| Microsoft Defender for Cloud regulatory compliance dashboard | Control Tower dashboard, Security Hub standards | CT shows control state, Security Hub shows security standard results |
| Management group "platform" and "landing zones" split | Security, Infrastructure and Workloads OUs | Similar idea; AWS keeps log and security accounts in the Security OU |
| Azure Blueprints (retired) | Landing zone updates plus IaC | Prefer IaC you own over a retired service as a lesson |
Architecture
What Control Tower puts in place:
The decision flow most banks actually follow:
Step-by-step implementation
This is the design you would carry out. In this course you do the "lite" column in the lab and stop short of enabling Control Tower.
Decide the home Region and governed Regions
For Harbourline: home Region eu-west-2 (London) with eu-west-1 (Ireland) as the paired recovery Region. Govern only these two (plus us-east-1 if global-service control-plane calls require it, which they do for some services). Deny everything else with the Region control or your own SCP. Write the reasoning down: it links to your UK GDPR data-location position (UK GDPR and data residency).
Choose the OU and account shape
Start from AWS Organizations structure: Security, Infrastructure, Workloads (Prod, NonProd), Sandbox, Suspended. Map them onto Control Tower: Security is required; the others you register as additional OUs.
Pick a controls baseline per OU
Mandatory controls are always on. For strongly recommended and elective controls, make a table: control, OU, reason, owner, whether it is preventive, detective or proactive. Start with a small set. Every detective control costs Config evaluations and creates alerts someone must triage.
Choose the account vending route
Console Account Factory for the first few accounts. If you manage dozens of accounts, put requests in Git and use AFT (Terraform) or Account Factory Customization (blueprints). See the Module 10 lesson.
Plan identity
Decide Identity Center groups and permission sets (read-only auditor, platform admin, developer, break-glass) before enabling. Connect Identity Center to the bank's identity provider.
Pilot in a throwaway organisation
Never turn Control Tower on in your production organisation first. Use a separate test organisation, run the whole lifecycle (enable, register an OU, enrol an account, trigger drift, update, decommission) and measure the cost for a month.
Document what it does not do
Network design, workload account baselines beyond the template, data protection, application logging, incident response and compliance evidence all remain yours. Put them on the plan.
AWS Console instructions
You will usually only read Control Tower in this course. If your employer's organisation has it, look around without changing anything:
- Console → AWS Control Tower → Dashboard. Shows the landing zone version, home Region, governed Regions, OU and account counts, and non-compliant resources.
- Organization lists OUs and accounts and their enrolment state.
- Controls library (or Control catalog) lists controls with their behaviour (preventive, detective, proactive) and guidance. Filter by framework or service.
- Account Factory shows the template and, if used, provisioned products.
- Landing zone settings shows version, drift and the option to update.
AWS CLI
# Is there a landing zone? (the home Region matters: run in the landing zone's home Region)
aws controltower list-landing-zones --region us-east-1 --profile learn-management
# Details of one landing zone: version, status, drift
aws controltower get-landing-zone --landing-zone-identifier <landing-zone-arn> \
--region us-east-1 --profile learn-management
# Browse the control catalogue (control IDs are ARNs; never guess one)
aws controlcatalog list-controls --region us-east-1 --profile learn-management \
--query "Controls[].{Arn:Arn,Name:Name,Behavior:Behavior}" --output table
# Which controls are enabled on an OU, and are any drifted or failed?
aws controltower list-enabled-controls --target-identifier arn:aws:organizations::<mgmt-acct>:ou/o-xxxxxxxxxx/ou-xxxx-xxxxxxxx \
--region us-east-1 --profile learn-management
# Enable a control on an OU (changes the real organisation: do not run in the course sandbox)
aws controltower enable-control --control-identifier <control-arn-from-catalog> \
--target-identifier <ou-arn> --region us-east-1 --profile learn-management⚠ VERIFY the command and argument names in your installed AWS CLI version (aws controltower help, aws controlcatalog help).
Infrastructure as Code
Enabling a control is declarative. Both examples are illustrative: do not apply them to a production organisation without a pilot.
using Amazon.CDK;
using Amazon.CDK.AWS.ControlTower;
using Constructs;
// Enable one control on one OU. Compiled against Amazon.CDK.Lib 2.277.0, not deployed. ⚠ VERIFY: property names in your version,
// and take the control ARN from `aws controlcatalog list-controls`, never from memory.
public class BaselineControlsStack : Stack
{
public BaselineControlsStack(Construct scope, string id, string ouArn, string controlArn) : base(scope, id)
{
_ = new CfnEnabledControl(this, "RegionDeny", new CfnEnabledControlProps
{
ControlIdentifier = controlArn,
TargetIdentifier = ouArn
});
}
}The lite equivalent, which you actually run in this course, is the SCP and OU Terraform in labs/m15-org-guardrails/infra/terraform. It gives you the preventive half of what Control Tower does, with every statement visible to you.
Complete code
A control inventory: for each top-level OU, how many controls are enabled and how many have failed or drifted. It compiles on .NET 10 against AWSSDK.ControlTower and AWSSDK.Organizations 4.1.x. It needs a landing zone to return data, so treat it as something to read now and run later.
dotnet new console -n ControlInventory
cd ControlInventory
dotnet add package AWSSDK.ControlTower
dotnet add package AWSSDK.Organizationsusing Amazon.ControlTower;
using Amazon.ControlTower.Model;
using Amazon.Organizations;
using Amazon.Organizations.Model;
// Control Tower control inventory: for each top-level OU, which controls are enabled and are any drifted or failed?
// Needs a landing zone. Run in the MANAGEMENT account, in the landing zone HOME Region:
// AWS_PROFILE=learn-management AWS_REGION=us-east-1 dotnet run
var org = new AmazonOrganizationsClient();
var ct = new AmazonControlTowerClient();
var organization = (await org.DescribeOrganizationAsync(new DescribeOrganizationRequest())).Organization!;
var root = (await org.ListRootsAsync(new ListRootsRequest())).Roots!.Single();
var ous = (await org.ListOrganizationalUnitsForParentAsync(
new ListOrganizationalUnitsForParentRequest { ParentId = root.Id })).OrganizationalUnits ?? [];
var problems = 0;
foreach (var ou in ous)
{
var enabled = new List<EnabledControlSummary>();
string? token = null;
do
{
var page = await ct.ListEnabledControlsAsync(new ListEnabledControlsRequest
{
TargetIdentifier = ou.Arn,
NextToken = token
});
enabled.AddRange(page.EnabledControls ?? []);
token = page.NextToken;
} while (token is not null);
var failed = enabled.Where(c => c.StatusSummary?.Status == EnablementStatus.FAILED).ToList();
var drifted = enabled.Where(c => c.DriftStatusSummary?.DriftStatus == DriftStatus.DRIFTED).ToList();
problems += failed.Count + drifted.Count;
Console.WriteLine($"{ou.Name,-16} controls: {enabled.Count,3} failed: {failed.Count} drifted: {drifted.Count}");
foreach (var c in failed.Concat(drifted))
Console.WriteLine($" ! {c.ControlIdentifier}");
}
Console.WriteLine($"Organisation {organization.Id}: {problems} control problem(s).");
return problems == 0 ? 0 : 1;Code explained line by line
- Two clients: Organizations for the OU list and Control Tower for the controls. Control Tower API calls must go to the landing zone's home Region, which is why the header comment sets
AWS_REGION. ListOrganizationalUnitsForParentreturns the OUs directly under the root. Control Tower controls are enabled per OU (the target identifier is the OU's ARN, not its ID).- The
do … whileloop followsNextToken, as in every paginated AWS API. StatusSummary.Status == FAILEDfinds controls whose enable or disable operation failed.DriftStatusSummary.DriftStatus == DRIFTEDfinds controls where someone changed what Control Tower manages. Both mean "go and look", not "everything is on fire".- The exit code lets a scheduled pipeline fail when any control has a problem. Pair it with an alert to the platform team.
- It never calls
EnableControl. An inventory tool should be read-only.
Security considerations
⚠ Learning shortcut — not for production
The lab builds a "lite" landing zone from OUs and SCPs only. It does not build central CloudTrail, Config aggregation, Account Factory, drift detection or the Audit/Log Archive accounts' access model; those are the next lessons and the capstone. A bank rolling out Control Tower would treat enabling it as a change to the production organisation: pilot organisation first, written risk assessment, rollback plan, change board approval.
- Do not edit Control Tower managed resources by hand (its SCPs, Config rules, IAM roles, StackSets). You create drift; fix drift by re-registering or repairing from the console.
- The Audit account is powerful. It can read across accounts. Few people, strong MFA, break-glass only for writes.
- Log Archive is the evidence store. Look at its bucket policy, encryption and retention, and add Object Lock if your records policy requires it (Centralised logging).
- Mapped is not compliant. A control "mapped to PCI DSS" helps your evidence. It is not a statement that you meet PCI DSS. Compliance is decided by the bank's compliance function and, for cards, the QSA.
- Preventive controls can break deployments. Test them in a Sandbox or Policy Staging OU first and check CloudTrail for denials.
Banking use case
Harbourline's platform team compares three options in a one-page decision record (fictional scores out of 5, higher is better):
| Criterion | Weight | Control Tower | Lite (own IaC) | Hybrid (CT + AFT) |
|---|---|---|---|---|
| Time to first compliant account | 3 | 5 | 2 | 4 |
| Fit with bank's target OU/account shape | 3 | 3 | 5 | 4 |
| Auditor-friendliness (explainable by reading code) | 2 | 3 | 5 | 4 |
| Ongoing engineering effort | 3 | 4 | 2 | 3 |
| Per-account base cost predictability | 2 | 2 | 4 | 3 |
| Skill in team today | 2 | 2 | 4 | 2 |
The weighted totals (score times weight, summed) are 50 for Control Tower, 53 for Lite and 51 for Hybrid, which is close to a tie. That is realistic: the decision is rarely a landslide. The team picks Lite for the first six accounts (low cost, learns every piece), commits to a three-month review, and keeps the OU names compatible so that registering them with Control Tower later is possible. The record names what would change the decision: more than ten accounts, a requirement for Control Tower's dashboard from the risk team, or the first audit finding about inconsistent baselines.
Common mistakes
- Enabling Control Tower in the production organisation as an experiment. It changes Organizations, Config and IAM across accounts. Pilot elsewhere.
- Choosing the home Region carelessly. Decide with residency and recovery in mind; assume it is hard to change.
- Enabling every elective control. Each adds Config cost and alerts. Start small and add with an owner.
- Editing managed resources by hand, then wondering why updates fail.
- Forgetting Regions. An ungoverned Region is an attack surface. Deny unused Regions.
- Treating Control Tower as the whole security programme. It is a baseline. The next lessons add central logging and central security tooling.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
ListEnabledControls returns AccessDenied or an empty list | Wrong Region (not the home Region) or not the management account | Use the home Region and the management account profile |
| Control shows Drifted | Someone changed a managed resource (for example an SCP attachment or an OU move) | Re-enable the control or re-register the OU as the console suggests |
| Landing zone update fails | Previous manual edits to managed resources | Read the failure reason; revert the manual change; retry |
| Enrol account fails | The account is missing a required role or already has a conflicting Config recorder | Follow the enrolment prerequisites in the user guide ⚠ VERIFY; remove the conflicting recorder |
| Unexpected Config cost after enabling | Detective controls evaluate every resource change in every governed account and Region | Reduce Regions and elective controls; review recorder settings |
| New account in the wrong OU | Vended with the wrong parent | Move it, then re-register or update the account |
Cost and cleanup
Lesson cost: free (design and read-only commands). The decision it informs is not free, so use the estimate steps above before enabling Control Tower anywhere.
If you did enable Control Tower in a throwaway organisation, decommissioning is a multi-step process and does not remove everything automatically ⚠ VERIFY current behaviour:
# 1. Disable controls you enabled
aws controltower disable-control --control-identifier <control-arn> --target-identifier <ou-arn> --region us-east-1 --profile learn-management
# 2. Delete the landing zone (this takes a long time and leaves accounts and some resources behind)
aws controltower delete-landing-zone --landing-zone-identifier <landing-zone-arn> --region us-east-1 --profile learn-management
# 3. Afterwards check for leftover Config recorders and S3 buckets in each account and Region
aws configservice describe-configuration-recorders --region eu-west-2 --profile learn-sandboxClosing the shared accounts has a 90-day post-closure period, and Config and S3 charges continue until you remove their resources.
Exercises
- 🟢 Guided: Fill in a decision matrix like the one above for Harbourline with your own weights. Explain which criterion would flip the result.
- 🟡 Challenge: Choose five controls you would enable first on the Workloads OU (one per theme: encryption, logging, networking, identity, Region). For each, say whether it is preventive, detective or proactive and what would break if it were wrong.
- 🔴 Live problem: The risk team says "We must use Control Tower, the regulator expects a managed landing zone." You are not aware of any such rule. Draft a reply that asks what exactly was requested, restates what Control Tower does and does not evidence, and proposes a way to meet the underlying need (consistent baselines, evidence of guardrails) in either design.
Assignment
Write docs/adr/0002-landing-zone-approach.md: the options, the weighted matrix, the decision, the review date, the "what would change our mind" list, and an estimate of the per-account monthly baseline cost for each option (use ⚠ VERIFY prices and show your arithmetic).
Solutions
The matrix in the Banking use case shows how little separates the options, so the weights matter more than the scores. If you weight "time to first compliant account" at 5 and "auditor-friendliness" at 1, Control Tower wins. If your bank must show auditors exactly how a guardrail is built, weight auditor-friendliness higher and the lite path wins. State the weights in the record and have the risk owner sign them.
One sensible set: (1) Deny the creation of unencrypted EBS volumes and S3 buckets without default encryption (proactive, blocks at CloudFormation time), (2) detect CloudTrail not enabled (detective), (3) disallow public access to RDS and S3 (preventive), (4) disallow IAM user creation except for named roles (preventive), (5) deny actions outside approved Regions (preventive). What breaks if wrong: the proactive rule blocks IaC that does not set encryption, so teams need template modules; the Region rule can block global-service calls if not written with an exemption list, as the lab's linter checks.
Ask for the source: which regulation or supervisory statement, or an internal policy? As far as the platform team knows, regulators ask for outcomes (segregation, logging, resilience, evidence) rather than a specific AWS product, but the compliance owner decides. Explain that Control Tower provides consistent baselines and a dashboard, not compliance. Offer: if the need is "consistent baselines with evidence", the lite landing zone with tested policies (the lab's hl-org validate output), an organisation trail and a Config aggregator produces equivalent evidence. If the need is the Control Tower dashboard, propose the hybrid.
Knowledge check
Answer: Service control policies (and resource control policies for supported services). Detective controls use AWS Config rules; proactive controls use CloudFormation Hooks.
Answer: No additional charge for Control Tower itself, but you pay for the services it uses and the controls it enables (Config, CloudTrail, S3 and so on). Detective controls appear as Config rule evaluations. Check the current pricing page.
Answer: Log Archive and Audit.
Answer: It creates drift. Landing zone updates and some account operations can fail until the drift is repaired.
Answer: No. Mapping means the control is relevant to a requirement family. Compliance depends on the whole environment, processes and an assessor's judgement.
Interview questions
- Control Tower or build your own landing zone? Depends on account count, fit with the target shape, explainability to auditors, team skills and running cost. I would pilot, estimate Config cost, and write the decision down with what would change it. A hybrid with AFT is common.
- What are the three control types? Preventive (SCP/RCP), detective (Config rules) and proactive (CloudFormation Hooks).
- What does Control Tower not give you? Network design, workload baselines beyond its template, application security, incident response and compliance evidence on its own.
- How would you reduce Control Tower's cost surprise? Govern fewer Regions, enable fewer elective detective controls, limit recorder scope where policy allows, and review the bill monthly.
Certification practice
A company is adding 40 accounts over the next year and needs consistent baselines, centralised logging, and a way for teams to request new accounts. The platform team is small and wants AWS to maintain the baseline. Which approach fits best?
A) Create each account manually and attach SCPs afterwards. B) Set up AWS Control Tower with Account Factory, and add customisations where needed. C) Use a single account with many VPCs. D) Write a custom CloudFormation StackSet only, with no organisational structure.
Answer: B. Control Tower provides managed baselines, shared logging accounts and account vending, which suits a small team with growing account counts. A scales poorly and invites inconsistency. C is not multi-account. D lacks the OU, logging account and guardrail structure.
A Control Tower landing zone shows an enabled preventive control as drifted after an engineer manually changed an SCP attachment. What is the correct response?
A) Ignore it; drift is cosmetic. B) Revert or repair the change and re-enable or reconcile the control through Control Tower. C) Delete the landing zone. D) Disable all other controls.
Answer: B. Drift means managed state no longer matches what Control Tower expects. Repair through Control Tower so updates keep working. A leaves the guardrail in an unknown state. C and D are disproportionate.
Which type of Control Tower control reports that an existing S3 bucket is unencrypted but does not block its creation?
A) Preventive B) Detective C) Proactive D) Mandatory
Answer: B. Detective controls use Config rules and report after the fact. Preventive blocks API calls, proactive blocks resource creation in CloudFormation, and "mandatory" is a guidance level, not a type.
Completion checklist
- I can list what Control Tower creates and name the two shared accounts
- I can explain preventive, detective and proactive controls and the guidance levels
- I filled in a weighted decision matrix and know what would change the answer
- I can estimate the per-account Config cost drivers
- I did not enable Control Tower in a real organisation
Key takeaways
- A landing zone is a structure, a baseline and a vending machine. Control Tower buys you one; IaC builds one.
- Control Tower has no service charge, but its Config and logging footprint scales with accounts and Regions.
- Controls: preventive (SCP/RCP), detective (Config), proactive (CloudFormation Hooks).
- Mapped controls help evidence; they never equal compliance.
- Decide with a written matrix, pilot first, and keep OU names compatible so you can change your mind.
Further reading
Independent learning resource — not affiliated with, endorsed or sponsored by Amazon Web Services. AWS and related marks are trademarks of Amazon.com, Inc. or its affiliates. All banking data in examples is fictional. Sample architectures are for learning and are not automatically compliant with GDPR, PCI DSS or any banking regulation.
AWS Organizations structure: OUs, the management account and policy inheritance
Design an OU tree for a bank, keep the management account nearly empty, attach policies so they inherit correctly, and check the whole design with a tested validator before you touch a real organisation.
Guardrails with SCPs and RCPs: deny by exception, and test before you attach
Design a small catalogue of organisation guardrails (Region limits, protecting audit and security services, a root-user deny, sandbox limits, a data perimeter), prove them offline with a tested evaluator and linter, and roll them out without locking yourself out.