Think of a hotel. A guest checks in, receives a key card, uses the gym and the breakfast room, and checks out three days later. Somewhere behind the desk, a system has to know that this guest exists, that the card in their hand really belongs to them, and that the card stops working when the stay ends. Now picture a hotel with four thousand rooms, hundreds of staff and a lot of robots delivering towels. That, in a nutshell, is the problem IdentityOps tries to tame in an IT department. Here is the idea, explained without the jargon.
The one-sentence version
IdentityOps is the routine upkeep of every account in your technology estate: creating it when someone arrives, adjusting it when their job changes, proving they are who they claim to be when they sign in, and turning it off, completely, when they go. The Ops Atlas, an online reference for operations disciplines, gives it a slightly more formal framing and openly says the term is still new, so different vendors describe it in different ways. Its page on the subject, the Atlas entry for IdentityOps, is long but readable if you like detail.
Who exists?
The first job is the guest list. For people, the starting point is the human resources record. When HR says someone has joined, an account appears; when HR says they have moved teams, their groups change; when HR says they have left, the shutdown begins. In the Atlas’s vocabulary this loop is joiner, mover, leaver. If your hotel only learned about checkouts when somebody remembered to tell the desk, rooms would stay unlocked for guests long gone. The same happens with accounts.
For machines, the guest list is trickier because there is no HR department for software. Scripts, scheduled jobs and integrations each carry their own credentials, which the Atlas calls non-human identities. Its advice is delightfully ordinary: start with a census, note an owner and a reason for each, hand out credentials with a short shelf life, swap them on a schedule, and retire a leftover in stages: turn it off, give it a grace period in case something breaks, then remove it for good.
How do they prove it?
The second job is proof. A key card is like a password: handy, and easy to lose or copy. Hotels add checks, such as a photo ID at the desk. Computers add a second step. But, as the Atlas points out, not every second step is equally sturdy. A code sent by text message, or a pop-up asking “Was this you? Tap yes”, can be tricked by a clever caller or worn down by repeated requests. Only physical security keys and smart cards are counted as truly resistant to this kind of trick.
Passwords themselves have had a rethink, too. Modern guidance cited in the Atlas favours long ones, fifteen characters or more if the password stands alone, with no strange composition rules, no forced changes every few months and a check against lists of leaked passwords. It is the “four random words” approach, in other words, and the nagging monthly change request is on its way out.
What happens when someone loses their key?
Every hotel has a story about someone who talked their way into a room that was not theirs. IT has the same weak spot at the help desk. The Atlas describes a ladder of three recovery routes. For ordinary cases, people recover through a second method they set up earlier. For trickier ones, the system releases a single-use code only after the person’s manager confirms them in person or on live video. For accounts with powerful rights, two administrators and supervised proof of identity are required. Ringing someone back on the number on file is called unreliable, because phone numbers can be stolen by swapping SIM cards. Whenever a recovery happens, the account owner should receive a notice, so a surprise reset is spotted quickly.
Why turning off the account is not enough
Back to the hotel. A guest checks out at 11 a.m., but they have already walked into the pool area. Cancelling the card does not remove them from the pool. In IT language, someone who is already signed in can carry on for a while after their account is disabled, because the system gave them a pass that is valid until it expires. According to the Atlas, the default life of such a pass can be an hour or so, stretching toward a full day in some configurations. So a good leaver process also revokes sessions and then checks that the revocation worked, since some tools take a few minutes and some applications keep their own sessions.
You can find a nice side-by-side of how that fits into the wider incident picture in our earlier article on handling incidents faster with AI-assisted operations, which shows how quickly small gaps turn into outages.
A tiny scoreboard
How do you know all this is working? By counting, carefully. The Atlas shares a neat example. Imagine forty people left the company. Thirty-six had their main account switched off on time, which sounds like 90 percent. But a trio from that group still had a working login in some smaller application, so only thirty-three leavers were truly finished, or 82.5 percent. The difference between the flattering number and the honest number is where the risk lives.
Two more easy-to-grasp scoreboard items from the Atlas:
- Of 1,400 people, 1,190 had a strong sign-in method registered, which is 85 percent. But only about two thirds of actual sign-ins used it, which tells you people can still choose a weaker door.
- Machine credentials older than a chosen limit, such as ninety days, should be flagged so someone can replace them.
The repeating chores
No hotel runs on one big annual clean. The Atlas suggests a calendar of small, regular checks: a daily comparison between the staff list and the accounts, a weekly spot-check of lost-key cases, a monthly list of machine credentials about to expire and a quarterly rehearsal where a pretend employee leaves and the team follows them through every system. A sweep for accounts that have been idle for three to six months rounds it out, with carve-outs for staff on sanctioned time away and for break-glass accounts that are supposed to sit unused.
What goes wrong in real life
The Atlas collects several widely reported incidents, and the lesson it draws is encouraging in a funny way. In each one, what failed was a process, not clever mathematics. A test system with no second step. A service login stored in somebody’s personal folder. Passwords that nobody had changed for years. A convincing phone call to a help desk. None of these needed a genius, which is why they can be fixed by ordinary teams with a checklist and a bit of patience.
If robots in the hotel picked your interest, our piece on securing fleets of robots and the risks involved shows how machine identities matter in physical settings too.
Is this only for big companies?
Not at all, though the Atlas is honest that a small organisation may not need a separate practice. A five-person company can do a very good version of this with a spreadsheet, a calendar reminder and a written recovery script. What it should avoid is buying an expensive governance product before its HR information is clean, because software cannot tidy a mess it has not been told about. It is also worth checking licences early: some features depend on specific paid tiers, and many online services keep their provisioning options for higher plans.
The takeaway
Keep a guest list that matches reality. Check that people are who they say they are, using methods that are hard to fake. Make recovery slower and safer than it feels convenient. Turn accounts off completely, and measure that they are off. Give every robot a name tag and an owner. That is IdentityOps in plain language, and the hotel analogy holds right to the end: the smoothest front desk is the one nobody notices, because nothing goes wrong.

