Book Appointment

A Practical IdentityOps Audit Checklist for Hospital IT Teams

Hospital IT

This is a working checklist for a hospital IT or administration group that wants to audit its own IdentityOps in an afternoon, then fix what it finds. No theory tour, no product pitch. It covers staff and system accounts only; it is not clinical or medical guidance, and nothing here describes any named hospital. Print it, assign owners, and tick items off with evidence rather than good intentions.

Before you start: set the scope in writing

The Atlas treats the field as still taking shape, with vendors drawing its limits in different places, so settle yours first. A sensible split puts account existence and proof of identity inside your audit. Permission design, access requests and the HR form itself sit outside it. Write two sentences stating what is in and what is out, and attach a date. You can read how the Ops Atlas entry on IdentityOps draws the same line if you want a reference point.

Section A: the roster

  • Is there one authoritative HR list of current staff, contractors and volunteers? Name who maintains it.
  • Does every active account trace to a person on that list, or to a registered machine identity with an owner?
  • Run a comparison today. Count accounts with no matching HR record. Anything above zero is a finding.
  • Are shared logins used by several humans? List each one and set a retirement date.

Section B: joiners and movers

  • When a person starts, does a baseline set of access arrive automatically with their role?
  • Is a second sign-in factor enrolled before first real use?
  • When someone changes department, does a trigger prompt a review so old duties stop carrying old rights?
  • Are applications that create users on first login, with no removal channel, flagged for a different control?

Section C: leavers, end to end

This is the section that earns the afternoon. Pull the last quarter’s departures and test a sample by hand.

  1. Record the time HR closed each person.
  2. Record the time the main account was disabled.
  3. Check whether open sessions and tokens were revoked, and when.
  4. Check every connected application for a surviving account, using the application’s own user export rather than the provisioning log.
  5. Compute two percentages: disabled within target, and fully gone everywhere.

The Atlas gives a worked example in which a directory-only figure of 90.0 percent turns into 82.5 percent once stray application accounts are counted. Expect your own gap to be visible too. Decide targets and write them down; the Atlas offers four hours for the main disable and twenty-four hours for every other tool as illustrations.

Section D: sessions and tokens

  • How long does a default access token live in your main sign-in system?
  • Does your leaver script call a revoke-sessions step, and has anyone tested that it works?
  • Do applications keep their own sessions that the central system cannot clear?
  • If an application uses OAuth connections for provisioning, do you know the order in which to pause and disable?

Section E: proofing and recovery

  • List the second-factor methods in use. Count SMS and tap-to-approve separately from hardware keys and smart cards.
  • Measure phishing-resistant coverage (how many people have one) and usage (how many sign-ins use it). The Atlas example shows 85.0 percent coverage alongside only 66.1 percent usage, meaning the policy still allowed weak methods.
  • Do administrators of the sign-in system use hardware keys? If not, fix that first.
  • Is help-desk recovery scripted, tiered, and notification-backed? Does it avoid phone call-back as the only check?
  • Do passwords follow current guidance: long minimum, a blocklist, no forced rotation?

Section F: machine identities

  • Is there a complete list of service accounts, keys, tokens and integrations?
  • Does every entry name its owner, state its reason for existing, carry a lapse date and say how frequently somebody revisits it?
  • What is the oldest credential, in days? Compare it to a ceiling such as ninety days.
  • When something is retired, is the sequence disable, wait, delete followed, with a record kept?
  • Do your cloud reports cover every kind of key? One cloud report the Atlas mentions looks at just two keys for each user and enumerates users while skipping roles, so blind spots are plausible.

Section G: dormant accounts

  • Is there a scheduled sweep for logins unused for ninety to a hundred eighty days?
  • Are approved leave, emergency accounts and long-cycle machine identities excluded by rule instead of by memory?

Section H: licences and plumbing

Cheap to check and often forgotten: automated deprovisioning may sit behind a higher licence tier in your identity platform or in a downstream application. List which tools have it and which do not. For the latter, add a manual step with an owner and a deadline.

Scoring and next moves

Give each section a red, amber or green mark. Do not buy a governance platform because section A was red; the Atlas cautions against purchasing one before the HR data is clean. Fix the data first, then automate.

A sensible ninety-day plan, adapted from the Atlas, runs like this. In month one, assign application owners, create the register, put hardware keys into the hands of every admin and sketch the lost-device procedure. In month two, connect one staff group to the HR feed, extend automatic removal to the ten most used applications and complete the machine-credential list. In month three, exercise session revocation, hold the first leaver drill and wire up alerts for password resets followed by changes to federation settings.

For wider context on how operations disciplines relate, a post here on the unified view of IT, data and cloud operations is a useful orientation, and a second on the basics of resilient IT operations pairs well with Section H.

Finish by booking the next audit before anyone leaves the room. Without a date, checklists become wall decoration, and a reminder in a shared calendar costs nothing. The first pass will be messy; the second shows whether the fixes held.