AWS Builder Center
One AWS account for everything? A startup's guide to untangling it

One AWS account for everything? A startup's guide to untangling it

A simplified guide for founders who already run on AWS: what a multi-account setup is, why it matters as you grow, what good looks like at each stage, and how to fix it when production and billing already share one account.

Almost every startup begins with one AWS account. You sign up, launch your app, and everything lives together: production, experiments, the billing, the root login. It works great for a demo. It quietly becomes a liability the moment you have paying customers, a second engineer, or your first SOC 2 question.
This guide walks through the whole picture in the order that actually helps:
  1. what multi-account even means
  1. why you should care
  1. how to do it
  1. what a good setup looks like as you grow
  1. what not to do
  1. and most importantly what to do if you're already in a mess.


Part 1: What are we even talking about?



An AWS account is a container. Everything you run, servers, databases, storage, lives inside one, and an account is the hardest security and billing wall AWS gives you. Think of each account as a separate room with its own lock.
A multi-account setup just means using several of these rooms instead of cramming everything into one. One for production, one for development, one that only holds the bill, and so on. You manage them together using AWS Organizations, which links all your accounts under a single umbrella at  no additional charge .
At the top of that umbrella sits the management account (also called the payer account). It receives the consolidated bill for everything and holds the top-level controls for the whole organization. Everything else is a member account hanging underneath it.
Figure 1 - A single shared account versus a small AWS Organization. Each account is a hard security

Figure 1 - A single shared account versus a small AWS Organization. Each account is a hard security and billing boundary.

Part 2: Why should you care?



This is not tidiness for its own sake. AWS's own  Well-Architected guidance  rates failing to separate workloads into accounts as a High risk - the same tier as leaving security holes open. Here is what that risk looks like in real founder terms.

1. Blast radius: One leaked key shouldn't take down everything

When your production database and a developer's sandbox share one account, a single leaked credential or one wrong IAM policy can reach everything. In a multi-account setup, if a developer database is compromised, only that account's resources are exposed. Other environments, logs, and security tooling stay untouched. AWS calls this limiting the scope of impact, and it is one of the headline benefits in the  multi-account whitepaper .

2. When someone leaves

If your whole company runs in one account, offboarding a departing engineer is nerve-wracking. Their access likely touched production, billing, and everything else, and you're never fully sure what a shared credential could still reach. With separate accounts and  central single sign-on  through IAM Identity Center, you revoke one identity in one place and their access to every account disappears at once. Federated logins also hand out  short-lived, auto-expiring credentials  instead of static keys that outlive the person.

3. Cost clarity: Who spent what

In one account, the only way to know whether prod or a dev experiment caused a spend spike is meticulous tagging on every resource, discipline almost no early team sustains (but ideally should). Separate accounts make cost tracking automatic, because an account is  the default unit AWS allocates costs to . And you don't lose volume discounts:  consolidated billing  still pools usage across all accounts.

4. Scale, Audits, and Quotas

AWS service quotas and API rate limits are  enforced per account . When dev and prod share one, a developer's load test can eat the API limit your live app depends on. And when regulated data shares an account with everything else, your entire account is dragged into audit scope, and isolating it in a dedicated account creates a clear data perimeter and shrinks what auditors examine.
The honest self-check
You probably don't need this on day one. AWS's own  guidance  says the need typically arises once you've found product-market fit, raised funding, and started hiring distinct engineering disciplines such as infrastructure, DevOps, or security. As a rule of thumb, if you can check three or more of these: (a) no containment when something breaks, (b) cost allocation is guesswork, (c) audit scope creep, (d) quota contention, (e) IAM sprawl, then it's time to plan the move.

Part 3: How do you actually do it?



There are four realistic paths, from "by hand" to "get help." Most startups blend the middle two.
Approach
Best when
Trade-off
By hand (AWS Console)
You have 2-4 accounts and just need to get off a single account fast.
Fine to start. Doesn't scale and is easy to do inconsistently.
AWS Control Tower (Guided setup)
You want a well-architected foundation without wiring it yourself.
 About an hour to set up 
. Free itself, but it turns on AWS Config in every enrolled account, which bills per item recorded - noticeable if you run lots of Spot or autoscaling. Easiest to adopt early, before there are hand-built accounts to enroll.
Scripts / IaC (Terraform / CloudFormation)
Your team already lives in infrastructure-as-code and wants repeatable account creation.
Most control and repeatability. automates compliant new accounts, but it runs on top of Control Tower.
AWS team or a partner (Get help)
You're mid-migration with live production and limited bandwidth.
Fastest path to "done right." Talk to your AWS account team or an AWS Partner.
The shortcut most startups should take
 AWS Control Tower  sets up the whole landing zone for you: a Security organizational unit with Log Archive and Audit accounts, baseline controls (guardrails), IAM Identity Center, and Account Factory so you can stamp out standardized new accounts on demand. If you expect to use it at all, turn it on early:  enrolling an existing account  means reviewing the baseline Control Tower deploys into it and clearing out any AWS Config recorder you already have.

Part 4: What good looks like (and how it grows)



There is no single right answer for every startup. AWS's own advice is to start small and layer in structure as your workloads grow. Here is the target picture at three stages - find where you are and aim one step ahead.

Seed / Early: 2 to 4 accounts

You have a live app and a handful of engineers
  • Management account: Billing and Organization only. No workloads, ever.
  • Production account: Your live, customer-facing app
  • Dev / Sandbox account: A safe place to experiment
  • IAM Identity Center: Turned on from day one (it  costs nothing extra ). One login per person across every account, short-lived credentials, one place to revoke access
  • (Ideally) Log Archive / Security account: one place all logs land
This is the minimum viable "good."  AWS's guidance  is blunt: don't run workloads in the management account.

Growth: Add structure and governance

Multiple teams, more workloads, first compliance questions
  • Group accounts into Organizational Units (OUs) by function: Security, Infrastructure, Workloads (prod + non-prod)
  • Deploy an AWS Control Tower landing zone if you haven't - it creates Log Archive + Audit accounts and applies baseline controls, your first guardrails without writing a policy yourself
  • Give every new workload its own account through Account Factory

Scale: Guardrails do the enforcing

Many teams, regulated data, dedicated platform/security people
  • Nested OUs by lifecycle (split dev / pre-prod / prod where policies differ)
  • Dedicated, immutable Log Archive + Security Tooling accounts
  • Service Control Policies (SCPs) as org-wide guardrails - broad near the root, strict near individual accounts. Test each one on an OU holding one or two accounts first; AWS  strongly advises against  attaching untested SCPs to the root.

Figure 2 - The same foundation, growing. Start at Seed and layer in structure only as you need it.

Figure 2 - The same foundation, growing. Start at Seed and layer in structure only as you need it.

Key Terms:

OU (Organizational Unit): Folder that groups accounts so you can apply one set of rules to all of them at once.
SCP (Service Control Policy): Org-wide guardrail that caps what an account is allowed to do (it limits, it never grants).
AWS Control Tower: Guided setup that builds, and then keeps governing, a well-architected multi-account environment for you.
IAM Identity Center: One login for everything: central single sign-on across every account, at no extra charge.

Part 5: What NOT to do / Anti-Patterns



These are the mistakes that look harmless early and get expensive later. If you recognize your setup here, don't panic, Part 6 is your fix.

1. Running production in your management (payer) account

The single most important one. Org-wide guardrails (SCPs)  do not apply to the management account , so anything you run there has zero guardrails - and if that account is compromised, the attacker gets control of your entire organization, not just one app. It also  complicates deploying Control Tower .

2. Production sharing one account with dev / test / staging

Separated only by naming and IAM policies, there are no hard walls. AWS  names this a common anti-pattern  directly - a compromised dev credential can reach production data.

3. Weak credential hygiene: root for daily work, no MFA, long-lived keys

The root user has unlimited, un-scopable power over the account and its billing. Use a normal admin identity for everyday work and put MFA (ideally a passkey or hardware key) on root. AWS  requires it for all account types  within 14 days of first sign-in.
Once you have an organization, turn on  centralized root access , which lets you remove root credentials from member accounts entirely; accounts created through Organizations already come without them.
Separately, a static access key pasted into a config file or CI system never expires on its own, and a key leaked in an old commit keeps working indefinitely. Use  short-lived, federated credentials  through IAM Identity Center instead.

4. Root email that belongs to one person

The root user's email address is the password-reset path for the whole account, and it is where AWS sends security alerts, abuse reports, and billing problems. If it is a founder's personal inbox or an individual work address, a vacation, a departure, or a lost phone cuts the company off from its own account. Use a distribution list your company controls (for example  aws-root-prod@yourcompany.com ): it forwards to several people, is used for nothing else, and you need one per account because every AWS account must have a unique email. AWS lists a group address among its  root user best practices . Do the same for the  alternate contacts  (billing, operations, security), which can also be distribution lists; in an organization you set them centrally for every member account from the Organizations console.

Part 6: "I'm already in a mess", now what?



This is the part most guides skip. You're not starting clean. You have live customers in a tangled setup. The good news: the fix almost never involves moving production. Find your starting point below.
Where you are
What to do
One account, everything in it, and you never created an organization
Scenario A, simple version - a new management account and one invitation.
Your management (payer) account also runs production, with other member accounts under it
Scenario A, full version - a new management account, then retire the old organization.
Several accounts, but prod shares one with staging or dev (and maybe the payer runs workloads too)
Scenario B - fix the payer first if needed, then split shared accounts one workload at a time.

Scenario A: the account that pays the bill also runs production

The counterintuitive fix

Do not try to "move production out" of this account. Instead, create a new, clean management account and bring your existing account in underneath it as a regular member. Production stays exactly where it is - no database migrations, no re-encryption, no DNS cutover. AWS's own  guidance  for teams who already run workloads in their first account says the same: create a new account for the organization, then migrate the old one in as a member.
What the move really involves. There is no "demote" button in AWS Organizations, so it helps to know what happens under the hood:
  • If your account is standalone (you never created an organization), this is the simple version: create a new account, create the organization from it, and  invite your existing account . Once it accepts, it's a member and you're done.
  • If your account is already the management account of an organization with other members, it has to leave that organization first, and a management account leaves only when  the organization is deleted  - which requires every other account to move out. Deletion is permanent: the organization's policies and all of its Cost Explorer data go with it, and IAM Identity Center, which needs an organization to operate, stops working for it.


Figure 3 - Scenario A: stand up a fresh, empty management account, then bring your existing account

Figure 3 - Scenario A: stand up a fresh, empty management account, then bring your existing account in beneath it as a regular production member. Production workloads never move - the account leaves its old organization (if it had one) and joins the new one.

The sequence at a glance (full version)

  1. Back up billing history (Cost and Usage Reports, Cost Explorer) to a bucket you own.  Deleting the old organization deletes it .
  1. Create the new management account: Dedicated root email on a distribution list, MFA on root, no workloads. Accounts migrating in from another organization  need it to be at least seven days old , so use the week for step 3. Set up Control Tower here if you plan to use it.
  1. Find what depends on the old organization - delegated administrators, services with trusted access, policies that reference your organization ID. The  Account Assessment for AWS Organizations  solution can help identify them. Make sure you can reach every account without Identity Center:  invited accounts  don't get  OrganizationAccountAccessRole  automatically.
  1. Move member accounts, non-production first. Since November 2025 an account  transfers directly : the new organization invites, the account accepts, no standalone step in between.
  1. Retire the old organization.  Delete it , invite the old payer into the new one, and accept. Then rebuild Identity Center (identity source, permission sets, assignments), delegated administrators, and Control Tower enrollment.
For the full runbook, including cleaning up legacy Control Tower stacks before you re-enroll, follow the  re:Post guide . It is written for larger environments, but the sequence is the same.

Scenario B: several accounts, workloads mixed together

Your payer may run a few things, and production shares an account with staging or dev. The target is a clean hierarchy: a dedicated management account and OUs by function. Here is the structure most startups end up with:
Root
|-- Security OU -> Log Archive, Audit accounts
|-- Infrastructure OU -> Networking (shared VPC, DNS)
|-- Workloads OU
| |-- Prod OU -> App1-Prod, App2-Prod
| `-- NonProd OU -> App1-Dev, App2-Staging
|-- Sandbox OU -> developer sandboxes
`-- Transitional OU -> accounts mid-migration (temporary)
The Transitional OU is a pattern from  AWS's own whitepaper : a temporary OU for housing accounts during a migration or restructuring, so they can be folded gradually into your standard governance structure. It's also a safe place to test guardrails on an account before its final move.
Step 0: if the payer runs workloads, do Scenario A first. Then split the shared accounts. The principle is Scenario A's: leave production where it is and move everything else out of its way, starting with the lowest-risk workload - a staging copy, an internal tool. For each workload that moves, the pattern is the same:
  1. Create the target account in Organizations (or through Account Factory if you're on Control Tower) and place it in the Transitional OU.
  1. Stand up the stateless layer from infrastructure-as-code. Compute, networking, IAM roles, queues in the new account. No IaC yet? This is the moment it pays for itself; don't click-ops a second copy you'll have to keep in sync.
  1. Move the state. RDS and Aurora:  share a manual snapshot  with the target account, copy it there, and restore from the copy; a shared encrypted snapshot can't be restored directly. Snapshots encrypted with the default KMS key can't be shared, so copy them to a customer managed key first and let the target account use that key. S3: replication or a batch copy. DynamoDB: export and import, or a cross-account backup copy. Rehearse with staging first.
  1. Grant cross-account access only where a dependency really exists - a resource policy or an assumable role - not blanket trust between accounts. Permanent cross-account links are the complexity you're trying to escape.
  1. Cut over. Point DNS at the new account - Route 53 weighted records let you shift traffic gradually - watch error rates, and keep the old copy running until you're confident.
  1. Decommission the old copy, then move the account out of the Transitional OU into its final home.

What breaks when a workload changes accounts (check before you move anything)

  • Hardcoded account IDs and ARNs in IAM policies, bucket policies, CI/CD config, application config.
  • KMS keys whose key policies trust only the old account.
  • Cross-account IAM roles and trust policies that name the old account.
  • Anything addressed by ARN from outside: SNS topics, SQS queues, SES verified identities, EventBridge buses.
  • DNS and certificates: a Route 53 hosted zone  is migrated  by creating a new zone in the target account, copying the records, and updating the name servers at your registrar. ACM certificates are requested again in the new account.
  • Network ranges: if the new account's VPC will connect to the old one, the CIDR ranges can't overlap.
  • Service quota increases you requested in the old account as new accounts start at the defaults.
The Multi-Account Transition Advisor (below) audits IAM roles and policies for hardcoded account IDs and ARNs, the first item on this list.

Do the low-risk thing first

You don't have to boil the ocean. The lowest-risk move is to create brand-new accounts for NEW workloads. There's nothing to relocate, so no migration risk. Save the bigger lift (splitting existing production, with its encryption and DNS-cutover care) for a deliberate, planned effort.
Gotchas worth knowing before you start
  • AWS credits (such as Activate credits) are  applied by calendar month , based on which organization, or standalone account, owned the account at the start of the month. A mid-month move can leave days uncovered or send credit to the old organization's bill. Move credit-holding accounts at the start of a month, and check credit sharing in Billing preferences afterwards.
  • The  default quota is 10 accounts  per organization, sometimes lower for brand-new ones. Request an increase early; pending invitations count against it.
  • Age and billing rules: an account created inside an organization  must be at least four days old  before it can be migrated, the destination organization must be at least seven days old, and the account's seller of record must match the destination's, or the move fails. Check the last one early if accounts were opened in different countries or came from an acquisition.

Let a tool do the audit for you

The  AWS Multi-Account Transition Advisor  is an agent from the AWS Startups prompt library that you run with your AI coding tool. It scans your resources, classifies them by environment, proposes how to split them across OUs, flags hardcoded account IDs and ARNs in IAM that would break after migration, and generates Terraform or CloudFormation to bootstrap your new organization. Give it read-only credentials while it scans, and review everything it generates before you apply it.

Talk to your AWS account team before you start

This is what your account team is for. If you have one (an account manager and a solutions architect), bring them in before you delete an organization or move an account - those steps are one-way. The useful ask is specific: "Here is our current account layout and the plan from this guide. What breaks?" They have watched other startups hit the gotchas above, can sanity-check your timing (the seven-day and four-day rules, credits at the month boundary, the account quota increase), and can bring in an AWS Partner if you want hands-on help with the migration itself. One expectation to set: account teams advise and unblock, they don't operate your accounts. No named team yet? Open a case with AWS Support - every paid plan includes technical guidance - or reach out; that's what we're here for.

Your move: Start small this week



You don't need a landing zone by Friday. You need momentum. The gap between "one shared account" and "a sound foundation" is crossed one deliberate step at a time - and the earlier you start, the cheaper it is. Migrating a tangled account later is far more painful than designing a little structure now.

Three things you can do this week

  1. Put MFA on your root user (a passkey or hardware key), move the root email to a distribution list, and stop using root for daily work. Fifteen minutes, huge risk reduction. Already have an organization? Turn on centralized root access too.
  1. Run the self-check. Count your warning signs - shared blast radius, guesswork cost, audit scope, quota contention, IAM sprawl. Three or more means it's time to plan.
  1. Create one new account. Put your next new workload in its own account instead of the shared one. Zero migration risk, and you start building the muscle.
Then: if you're on a single account, run the Multi-Account Transition Advisor and take its output to your AWS account team before you change anything.

Go deeper



The essentials, if you want to read the source material:
Written for AWS startup customers. Always verify AWS service behavior and limits against official documentation before making production changes.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article