Anyone who flies or maintains light sport aircraft has met the trade-off between a rule written for everyone and a rule tuned for one situation. A club-wide policy is easy to explain and enforce, but it can be clumsy for a single aircraft or a single pilot. Cloud operations faces the same choice. This comparison looks at three decisions that people running cloud estates have to make, namely where to apply a restriction, whether to enforce or only observe, and how many accounts to keep, using the CloudOps entry of the Ops Atlas as the factual backbone. No theory for its own sake, just what each option gives you and what it costs.
Decision one: apply a rule everywhere, or by group?
The Atlas uses Region restrictions as its example, and it names two controls in AWS Control Tower. One, AWS-GR_REGION_DENY, operates across the entire landing zone. The other, CT.MULTISERVICE.PV.1, is attached to an organisational unit, which is a branch of the account tree. Here is how they compare in everyday use.
Landing-zone-wide control
- Strength: one switch, one list of allowed Regions, no gaps between teams.
- Weakness: every team lives with the same list, even if a research group has a legitimate reason to work somewhere else.
- Suits: small organisations or those with a uniform data-residency requirement.
Per-unit control
- Strength: you can allow a wider list for a sandbox branch and a tighter list for production.
- Weakness: more settings to review, and an easy place for drift. A new branch created without the control is simply unprotected.
- Suits: larger estates with different risk profiles.
The aircraft parallel is a club rule that every aircraft needs a daily fuel-sample check, versus a rule that adds extra items only for the aircraft used for long cross-country flights. The first is simple. The second fits the job better but needs more care to keep up. If you want a concrete picture of how focused checklists operate in this world, the rundown of fuel-system slips light sport pilots ought to avoid shows how a narrow list can catch narrow problems.
Decision two: enforce straight away, or observe first?
Whichever scope you choose, you must also pick a mode. The thirty-sixty-ninety outline in the Atlas has guardrails going live in audit mode around the second month, with enforcement arriving by organisational unit after that.
Audit mode
- Strength: no one gets blocked, and you learn how big the problem is before you change anything.
- Weakness: nothing actually stops a violation, and reports can pile up unread.
Enforce mode
- Strength: the rule holds. A denied Region cannot be used by accident.
- Weakness: surprises. The Atlas lists blind spots: resources that already exist may become unmanageable once their Region is denied, global services such as identity and certain us-east-1 endpoints fall outside regional denial, and on another platform the location policy skips resource groups and the global pseudo-location. A third vendor’s constraint covers creation but not the places where data is processed.
The Atlas’s advice amounts to doing both, in order: observe, fix the findings, then enforce, and always keep an exception route. A rule with no exit path pushes people to work around it, which is worse than having no rule. For a pilot, this is the difference between announcing new minimums with a month of logged feedback and grounding the fleet on Monday morning.
Decision three: few accounts or many?
The third decision is about how finely to divide the estate. The Atlas frames the trade this way: splitting the estate contains the fallout from an error and hands every account its own allowances, while multiplying the baseline set-up you must keep current.
One large account
- Strength: fast to start, simple to explain.
- Weakness: shared limits, a shared bill and mistakes that reach everything. The Atlas names this among its recognised anti-patterns.
Many purpose-built accounts
- Strength: contained failures and clear cost ownership.
- Weakness: more set-up work, and a need for a repeatable way to create accounts. A further detail from the Atlas: the organisation-level restriction policies skip the management account itself, and the vendor’s own guidance is to keep workloads out of it.
Compare it with owning one hangar for every aircraft, tool and fuel drum versus several smaller buildings. The single building is cheaper to rent and awful if there is a fire. Of the three decisions, this one is hardest to reverse, so make it deliberately. A good companion read for the planning mindset is the checklist approach to landing gear for light sport owners, where the habit of listing items before acting is the whole point.
How to run the comparison in your own estate
Before choosing, collect three facts. How many accounts exist today, and who owns each? Which Regions are in actual use, as opposed to officially allowed? And what happened the last time a rule blocked somebody? Those answers usually point to the right option faster than any debate. A short workshop with the people who will live with the rule, including whoever handles the invoices, avoids the common outcome in which a tidy policy is written by one group and quietly ignored by another.
Also decide how you will measure success. The Atlas proposes tracking Region conformance and baseline coverage, with a typical target of 95 percent for the latter. Pick one number per decision, record it before the change and again a month afterwards. If the figure does not move, the rule is decoration.
Side-by-side summary
- Scope: wide is simple and rigid, per-unit is flexible and harder to keep consistent.
- Mode: audit teaches, enforce protects. Use them in sequence.
- Accounts: few is cheap and fragile, many is tidy and demanding.
A fourth comparison worth a paragraph: one provider or several
Teams often ask whether a second provider is the cure for these trade-offs. The Atlas answers cautiously. Bringing in a second provider means a duplicate management interface, a duplicate sign-in system and a duplicate rulebook, so every decision above has to be made twice. The Atlas repeats a vendor’s warning that the added intricacy could cancel out initial gains. Multi-cloud can still be justified, for instance by regulation or a specific capability, but it is not resilience by default. Teams in markets where cloud skills are scarce may find the discussion in this roadmap on infrastructure skills and training useful when sizing the people cost.
Mistakes people make when choosing
Three errors recur. The first is copying another organisation’s settings without checking whether your accounts, Regions and teams look anything like theirs. The second is turning on enforcement without telling anyone, which produces a flurry of urgent tickets and a reputation for the operations team as obstructive. The third is treating the decision as permanent. Rules should be reviewed on a schedule, because the estate grows, staff change and providers add new features that make older workarounds unnecessary. Put a review date on every guardrail the day you create it, and write down who is allowed to grant an exception.
Verdict
For a small organisation that is just starting: pick the landing-zone-wide control, run it in audit mode for a month, then enforce, and split production from everything else into at least two accounts. For a larger estate with varied needs: use per-unit controls, but automate their attachment so that a new branch is never created bare. In both cases, read the Ops Atlas page on CloudOps for the full list of caveats, and remember that it describes the field as emerging. Test each choice on a small branch before rolling it out, record who owns each exception, and review the lot every quarter.

