List the accounts that actually matter

Start with the client systems that hold customer data, money movement, or production controls. List every contractor account that can read, change, approve, or export in those systems. Record the system, the account, the role, and the last login. A review that starts from a list of people rather than systems will miss shared and service accounts.

Include accounts that belong to the provider rather than the client. An agency admin account can reach several client tenants and is often overlooked. Record the owner, the tenant scope, and the business reason. The review should cover every path into the client's data.

Mark the accounts that are shared by a team, such as a queue mailbox or a reporting login. Shared accounts need a named human owner even when the password is in a vault. Record the owner and the people who may retrieve it. A shared account with no owner cannot be reviewed.

Add service accounts and integrations that run without a person. Record the purpose, the data they touch, and the person who can revoke them. A service account that nobody remembers is a standing risk. Include it in the review even when it has no interactive login.

Set the review scope with the client before the first cycle. Name the systems in scope and the systems that are explicitly out. Record the client sign-off and the date. An unclear scope produces arguments about coverage instead of decisions about access.

Set a cadence by risk, not by calendar habit

Rank each account by the harm that would follow a misuse. Money movement and production change usually rank higher than a read-only report. Record the rank and one sentence of reasoning. The rank decides how often the account is reviewed.

Review the highest-risk accounts monthly and the rest quarterly. A quarterly cycle is too slow for an account that can approve refunds, while a monthly cycle is wasteful for a dashboard login. Record the cadence beside the rank. The cadence should follow the risk, not the convenience of the reviewer.

Trigger an extra review when a person changes role, leaves the project, or moves to another client. The calendar will not catch a departure in time. Record the trigger and the person who must report it. The provider and the client both need a way to raise the event.

Review service accounts whenever the integration changes or the vendor changes ownership. Record the change and the review date. A service account can outlive the project that created it by years. A review that skips them leaves a standing key in the client environment.

Put the next review date in the account record, not only in a spreadsheet. A date that lives in a personal calendar disappears when the owner is away. Record the owner and the backup. The review should survive a single person's vacation.

Run the review with evidence

Ask for the business reason before looking at the permission list. An account without a current reason should be a candidate for removal, even if the permissions look normal. Record the reason and the person who confirmed it. A review that starts with permissions tends to preserve the status quo.

Check the last login, the last action, and the access grant date for each account. A dormant account with broad access is a different finding from an active account with the right role. Record all three dates. The combination tells the reviewer whether the account is used and whether the access is still needed.

Compare the granted role with the role described in the work instruction. A person may need read access but hold edit rights from an earlier project. Record the gap and the requested change. The review should test the fit between the job and the permission.

Ask the client to confirm the accounts that touch their data, not only the provider manager. The client may know about a vendor or a shared login the provider has forgotten. Record the client confirmation and any new account discovered. Two-sided confirmation is stronger than a provider-only review.

Record the evidence you used, such as an export, a screenshot, or a ticket reference. A review note that says checked is not evidence. Store the evidence with the review record. The next reviewer should be able to repeat the check without asking.

Remove access on a dated path

Decide the removal date when the review identifies an account that is no longer needed. A finding without a date becomes a permanent exception. Record the date, the owner, and the system where the removal happens. The date should be close enough to matter and realistic enough to complete.

Remove the access in the source system first, then update the inventory. A spreadsheet marked revoked while the account still works is a false control. Record the system confirmation and the time. The inventory should follow the system, not the other way around.

Check the removal from the user's side, not only the admin screen. Ask the former user or a test account to attempt the access and record the result. A cached session or a group membership can keep the door open. The test is the proof that the removal worked.

Handle shared and service accounts with a rotation and an owner update. Removing one person from a shared account is not enough if the credential stays the same. Record the rotation, the new owner, and the people who received it. The account should not remain a standing key to the client's data.

Tell the client when a high-risk account is removed and when a replacement is created. A silent change can break a workflow or leave the client unsure who has access. Record the notice and the effective time. The client should never learn about an access change from a failed task.

Report gaps without turning the review into blame

Report the count of accounts reviewed, removed, and still waiting for a decision. Add the number of accounts with no business reason and the number past their review date. Record the counts and the scope. A few clear numbers show whether the cadence is working.

Group findings by system and by cause, such as a missing offboarding step or a role that was copied from an old project. Record one example per group without naming individuals in a public report. The cause points to a process fix, while a name points to a person. The cause is what the next fix can actually change.

Give the client and the provider manager the same summary. A one-sided view hides the accounts each party owns. Record the recipients and the date. The review should produce a shared list of actions, not a private scorecard.

Follow up on the removals that missed their date and record why. A missed removal may mean the system owner was unavailable or the dependency was not understood. Record the reason and the new date. The follow-up is where the control becomes real.

Review the cadence itself once a quarter. Ask whether the high-risk list is still right, whether the evidence is easy to collect, and whether removals happen on time. Record one change and an owner. A cadence that is never reviewed becomes a ritual.

For the next step, compare this workflow with Audit delegated mailbox permissions after a client contact leaves and Set a document retention rule for outsourced support records. For access and oversight, use the authoritative planning reference alongside your own policy.