Imagine a small charter brokerage whose dispatch desk never sleeps. We will call the team the Harbour Desk, and everything about them is invented for the sake of the story. The Harbour Desk books aircraft, arranges ground transfers, sends passenger manifests to handlers and keeps quotes in a shared workspace. One Monday, a scheduling tool they barely remembered adding turned out to hold the power to read every manifest. Nothing bad happened. But the scare started a conversation about what the Ops Atlas calls IAMOps, and how a team like theirs might handle it.
The Monday that started it
The operations lead in our story was reviewing a vendor renewal when she noticed that a calendar add-on carried the same level of access as a senior coordinator. Nobody had decided that. The add-on had been connected in a hurry during a busy holiday week, with whatever setting made the error message go away. Everyone on the team had a version of the same memory: “It was easier to say yes to everything and sort it out later.”
That sentence is the first myth the Atlas addresses. It observes that once an over-broad role is running, usage data will only ever show it working, so the broad grant seems to be confirmed by experience. The tightening that was meant for later quietly never comes.
Reading the rulebook out loud
The team started by borrowing a framing from the Atlas page on IAMOps, which presents the subject as authorisation handled like software: rules written down, tested and gradually improved. The page admits that the term is only an informal label and no formal field, which suited a team with no appetite for jargon.
Their first exercise was a whiteboard session. For each sensitive action, such as exporting manifests or editing a quote, they wrote four things: who asks, what they do, which record is touched, and under which conditions. The dispatcher with the longest tenure spotted that the conditions column was blank on every line. Anyone could do anything at any hour, from any device.
Grants, ceilings and a surprise
The second discovery was about structure. The brokerage’s workspace had an account-wide setting that supposedly “allowed” exports only for managers. The newest hire pointed out that a ceiling like that never gives anyone a right; it only narrows what other settings have already granted. The add-on’s own role was the thing granting access, and the account-wide setting did not touch it because the add-on was not a person. The Atlas draws exactly this line between rules that grant and rules that cap, and adds that an explicit deny overrides everything else.
They fixed the oversight in the plainest way possible: they removed the add-on’s manifest permission, kept its calendar permission, and wrote down the owner’s name beside the integration.
Measuring instead of guessing
A week later the team asked how many other roles carried unused abilities. They adopted the measure the Atlas demonstrates, permission utilisation, which sets a role’s full list of allowed actions beside the list it really exercised. The Atlas example has one role using 27 of 180 actions, a 15 percent score. The Harbour Desk did not have thousands of roles, so they could simply read the list by hand each month and ask the owner about every ability that looked idle.
They also learned not to trust the reports blindly. Quarterly tasks, such as the seasonal fleet-availability export, did not appear in a one-month window. The longest-serving dispatcher nearly deleted that ability before somebody remembered the schedule.
Retiring the spare keys
The final piece was credentials. The accounting connector used a static key created more than a year earlier. The Atlas quotes an industry survey in which a majority of cloud identities had such aged keys, and recommends short-lived, automatically issued access in their place. The team could not federate everything at once, so they followed the order the Atlas suggests: the most critical pipeline first, the rest over later months.
What changed in the culture
Perhaps the biggest shift had nothing to do with tooling. Requests for more access now arrived with a sentence explaining why and for how long. Approvals took a little longer, which the Atlas treats as a reasonable price, citing an example where the wait rose from 36 to 48 hours while the median share of permissions in active use rose from 22 to 41 percent. That trade will look familiar to anyone who has watched a premium service accept a slower booking flow for the sake of discretion; our article on the preferences luxury travellers bring to a charter booking explores similar trade-offs between speed and care.
The transfer partners got the same treatment. Ground handlers only needed the pickup time and passenger count, not the full manifest, much like the arrangement described in our article about why private car services work well for airport transfers, where the driver receives only what the journey requires.
Lessons the Harbour Desk wrote on the wall
- Every integration has a named owner and a reason.
- Grants are narrow first; widening needs a note.
- Idle-looking permissions are questioned, not silently dropped.
- Long-lived keys are on a retirement schedule.
- A short monthly review beats a heroic annual one.
Stories like this are invented, but the shape is familiar to anyone who has run a busy operation. The useful part is the order: inventory, then ownership, then measurement, then removal. Try it on one integration this week, and see what the whiteboard tells you.

