How to Set Up a Daily Leaver Check in Seven Steps

How to Set Up a Daily Leaver Check in Seven Steps

One control from the PeopleOps toolbox pays for itself faster than almost any other: a daily comparison between the people record and the live accounts. It sounds modest, and it is. Yet the Ops Atlas lists the gap between “ended in HR” and “still active in the account system” among the most common operational failures. This guide shows how to build the check in seven steps, with decisions explained along the way. The framing comes from the Atlas reference on PeopleOps, and the engineering choices are ours.

Before you begin: what you need

  • Read access to a list of people with their employment status and end date.
  • Read access to a list of accounts in your identity system, with status and last change time.
  • A place to send alerts that someone will actually read.
  • One named owner for the result.

You do not need a large project. A script and a scheduled job are enough to prove the idea, and you can replace them later.

Step 1: Choose the join key

Everything depends on matching a person in the first list to the same person in the second. The Atlas is firm that the join must use a permanent worker identifier, not an email address, since mailbox names change after marriages, name corrections and rehires. If your identity system does not store that identifier, add it as an attribute first. Skipping this step produces false alarms that train everyone to ignore the report.

Step 2: Define what “gone” means

Decide on one rule for when someone counts as a leaver. The usual answer is that the effective end date and time have passed. Because the Atlas recommends entering terminations ahead of time with an effective moment, the check can also look forward and list people due to leave in the next day, giving system owners notice.

Write the rule in one sentence and put it in the job’s comments. Ambiguity here is the root of arguments later.

Step 3: Pull both lists the same way every day

Schedule the extraction at a fixed time and record the timestamp. Prefer a direct query or an API call over a file dropped on a server, since the Atlas warns that fragile flat-file transfers and scheduled pulls can make terminations lag. If you must use files, add a check that confirms the file is fresh and complete, for example by comparing the row count against yesterday’s.

Step 4: Compare and classify

Compute three groups.

  1. Ended but active: people whose end time has passed while an account is still enabled. This is the group that matters most.
  2. Active but unknown: accounts with no matching person, often orphaned logins, service accounts or contractors handled outside the staff process.
  3. Known but missing: current staff who should have an account and do not, a sign that onboarding stalled.

Keep a fourth, small bucket for accounts you have deliberately exempted, such as shared mailboxes, each tied to a responsible person and a date for taking another look.

Step 5: Alert, do not just report

A line in a spreadsheet that nobody opens is not a control. The Atlas says that “ended in HR, active in identity” must raise an alert. Send each ended-but-active account straight to the system owner with the leaver’s end time and the promised deadline. Escalate to a manager if the item is still open at the halfway point.

For a rehire, check that the identity was reactivated with no prior access. Restoring an old permission set quietly undoes everything your previous leaver process achieved.

Step 6: Start a clock and measure it

Revocation latency is the time between the end of employment and the moment of disablement. Choose a target before you look at the data. The Atlas gives sample targets such as disabling access within an hour for at least 99 percent of leavers, but it is clear those are examples. A published US control family leaves the timing to the organisation, so your promise is your own to set. A documented company case allows five working days, with critical systems inside 24 hours. Pick a figure suited to your risk.

Record the latency for every leaver in a simple table. After a month you will have a distribution, and the slowest ten percent will tell you which systems need attention.

Step 7: Close the blind spots

The first run will look great and mislead you, because your identity system only knows about the tools connected to it. Make a list of everything that sits outside single sign-on, then add each to the check or assign a person to review it. Include shared credentials, which need rotating when a holder leaves, and automation bots owned by a departing person.

Finally, run a same-day sanity test now and then: the list of terminations, the list of leavers and the list of disabled accounts should all agree. A disagreement points to a flaw in the process itself, rather than a quirk of the data.

A worked picture of one day

Imagine the job runs at six in the morning. It finds twelve people marked as ended and two enabled accounts among them. One is a contractor whose sponsor forgot to update the end date; the other is a developer whose account is attached to a build bot. The alert goes to two different owners, each with the leaver’s end time and a deadline. By noon the first is disabled and the second has been reassigned to the developer’s team lead, with its keys rotated. The latency table gets two new rows, and the weekly review sees that both delays came from systems outside single sign-on, which is the cue for step seven.

That small story shows why the check works. It does not depend on anyone remembering. It depends on a clock, a rule and a name.

Common mistakes while building it

  • Matching on email and then drowning in false positives.
  • Running the check weekly. By then a mismatch has been live for days.
  • Alerting a shared inbox rather than a named owner.
  • Counting a disabled account as done while its API keys still work.
  • Treating the metric as a performance score for the person who disables accounts, which tempts them to hide delays.

Next improvements

Once the daily check is stable, add the mover case: when someone changes role, list their accumulated entitlements for review. Then consider event-driven provisioning so that a termination in the people system pushes a change to connected tools straight away.

For wider operations reading, see our overview on how operations teams are using AIOps day to day“>how operations teams are using AIOps day to day and the walkthrough of setting up delivery pipelines for software teams“>setting up delivery pipelines for software teams. Both share the idea that small, automated, daily checks beat heroic clean-ups.

Seven steps, one owner, one daily alert. Together they convert a fuzzy concern into a figure that you can push down month after month.