When the same item reaches the safety review board for the third quarter running, the problem is not the agenda. It is that the decision has been placed at a level where nobody is able to take it, or at two levels at once, so that each defers politely to the other. The split between a review board and an action group is not an organisation chart exercise. It is a rule about who may spend, who may accept, and who may only recommend — and it either exists in writing or it does not exist at all.
Two bodies the regulation never actually requires
Start with the honest position, because it changes how the rest is read. The binding text does not order you to hold two meetings. ICAO Annex 19 requires a safety management system with defined components; the safety management manual that accompanies it describes a safety review board and safety action groups as a workable structure. On the European side, the management-system requirement in the annex that covers your approval — the air operator organisation requirements, Part-145 for maintenance, the service provider requirements for an air navigation service — obliges you to identify hazards, assess and mitigate risk, monitor performance, manage change and review the whole thing. The safety review board chaired by the accountable manager, and the action groups reporting to it, live in the acceptable means of compliance rather than in the rule.
Two things follow. First, an inspector audits functions, not furniture: whether risk was assessed, whether somebody with the authority to spend money accepted the residual, whether actions were verified. Second, if you do adopt the two-body model — and for anything above a handful of people it is the least bad one available — the burden is on you to write down where the boundary sits. A model borrowed without its boundary is worse than no model, because it manufactures the ambiguity it was meant to remove.
| Safety review board | Safety action group | |
|---|---|---|
| Question it answers | Are we safe enough, and are we resourcing it? | Is this hazard controlled, and by when? |
| Chair | The accountable manager, in person | The manager of the operational area concerned |
| Horizon | The period ahead: objectives, indicators, changes, budget | The period at hand: reports, assessments, actions in flight |
| Authority over resources | Can commit them | Can spend what has been delegated, and no more |
| Authority over risk | Can accept a residual risk on behalf of the organisation | Can control a risk; cannot accept one it has not been delegated |
| Normal output | Decisions, resources, direction, and closure on escalated items | Assessments, actions with owners and dates, and escalations |
| Typical failure | Reviewing individual occurrences | Quietly accepting a risk above its authority |
The test that separates them
Every workable boundary reduces to four questions asked of a proposed decision. Does it commit resources beyond the group’s delegated ceiling? Does it change a policy, a limit, a procedure that other departments depend on, or a published objective? Does it accept a residual risk that the matrix places outside the tolerable band? Does its consequence land outside the area the group manages? One yes sends the item up. Four noes keep it down, and keeping it down is not a discourtesy to the board — it is the reason the board has time to do its own job.
The most common design error is to write the boundary in terms of importance. Importance is unfalsifiable and everything becomes important eventually. Write it in terms of authority: a monetary ceiling, a named set of documents that only the board may amend, a defined region of the risk matrix, and a list of functions outside the group’s scope. Those four are testable by somebody who was not in the room, which is the only property that matters at two in the afternoon eight months later.
The board and the action group have different jobs. The board decides big things like budgets and accepting risks. The action group handles day-to-day hazards. If you don’t write down who does what, people will get confused and nothing will get done.
An airline’s action group identifies a runway excursion risk at a short runway (e.g., 1,500 meters) in high temperatures (e.g., 35°C). They propose training adjustments, but the board must approve it because it affects the entire fleet’s operations and requires budget allocation.
What sends an item up, and what keeps it down
| Trigger | The test | Who decides |
|---|---|---|
| Residual risk outside the tolerable band | Where the assessment lands on the matrix after the proposed mitigation, not before | Board — acceptance of residual risk is an act of the accountable manager |
| Cost or resource above the delegated ceiling | A figure written in the safety management manual, not a judgement | Board |
| Change to a controlled document or a published limit | Is the document on the list only the board may amend? | Board |
| Consequence crossing departments | Does the mitigation require somebody outside the group to do something? | Board, or a joint group convened for the purpose |
| Safety performance indicator through its alert level | The alert level itself, computed in advance, not the feeling in the room | Board — it owns the indicators |
| Action overdue by a defined number of cycles | A counter, not a discussion | Board, on the overdue action only |
| Recurrence after an action was closed | The same hazard reappears after verification said it was controlled | Board — the failure is in the process, not the hazard |
| Everything else | None of the above applies | Group, and the board sees the summary line only |
The last row is the one that saves the structure. If the board’s papers contain every assessment the group made, the board will read none of them, and the escalated items will drown among the routine ones. Send up the exceptions and the count of everything else.
The three ways it goes wrong
- The first failure is a board doing group work. The symptom is an agenda made of individual occurrences — a runway excursion, a tooling error, a passenger event — each discussed on its merits by people who were not there and cannot act. The board is at that point the largest and most expensive investigation team in the company, and its actual job, which is to look at whether the portfolio of risk is moving in the right direction, has fallen off the end of the agenda.
- The second is a group doing board work. It is quieter and more dangerous. A group assesses a hazard, cannot control it within its means, and concludes that the residual risk is acceptable in the circumstances. Nobody outside the room has agreed to that, nobody with the authority to fund an alternative has seen it, and the risk register now records an acceptance that was never made. When this surfaces it usually surfaces in an audit finding, and it is almost never bad faith — it is a group behaving reasonably in the absence of a written ceiling.
- The third is the double treatment. The same subject appears in both sets of minutes, each recording that the other body is looking at it. Nothing is decided, and the item acquires seniority through repetition until somebody proposes an action large enough to be visible and small enough to be affordable. The tell is textual: search two consecutive sets of minutes for the same noun phrase, and count how many times it appears without an owner and a date attached.
The trap. A meeting that treats « issue a notice to crews » or « cover it in recurrent training » as an action. Those sit at the weak end of the hierarchy of controls, below elimination, substitution and engineered defences, and they transfer the burden of the defect to the person least able to carry it. They are legitimate as an interim measure with an expiry date. They are not a mitigation, and a residual risk recalculated on the strength of a memo is a residual risk that has not moved.
The action, not the discussion, is the unit
Whatever the governance shape, the object that survives the meeting is the action, and most actions are written in a way that cannot be closed honestly. A closable action has four parts, and the fourth is the one that is routinely missing.
| Part | The weak version | The version that survives an audit |
|---|---|---|
| Owner | A department, or « operations » | One named person, who was in the room or has agreed since |
| Date | « Next review » | A calendar date, set against the risk and not the convenience |
| Completion criterion | « Review the procedure » | The artefact that will exist: revision X of the manual, issued and distributed |
| Verification of effectiveness | Absent | What will be measured, when, and what result would reopen the action |
The distinction that the fourth line protects is between an action closed and a risk controlled. They are not the same event and they rarely happen on the same day. An amended procedure exists the moment it is published; whether it changed behaviour is a question that can only be answered later, by an audit sample, a flight data query, a report count or an observation programme. A system that closes actions on publication and never returns is a system that has confused activity with control, and it will keep meeting the same hazards under new names.
Alert levels: the mechanism that decides what the board hears
Escalation by indicator only works if the threshold was set before the data arrived. The usual method is to derive an alert level from the previous period’s outcomes — a mean, plus one or more standard deviations, computed on a rate rather than a count so that traffic variation does not create phantom trends — and to set a target level separately as the improvement the organisation intends. The two are different objects and confusing them is common: the alert level says something has changed, and the target says something should change.
Set that way, the indicator becomes a trigger rather than a topic. The board does not debate whether a figure feels high; it observes that a level computed months ago has been crossed and asks the group for an assessment. That single discipline removes most of the negotiation from the agenda, and it is also what makes an indicator worth collecting at all — an indicator with no alert level attached generates a chart and no decisions.
Two agendas that do not overlap
Build the board’s agenda from exceptions and the group’s from work. The board takes safety performance against alert and target levels, the escalated risks and only those, actions overdue past their threshold, the management-of-change items for anything the organisation is about to do differently, resources and the decisions only the accountable manager can take, a sampled look at the effectiveness verification of items closed earlier, and the outcomes of audits and oversight. The group takes the reports that arrived, the assessments in progress, the actions in flight with their owners and dates, its own local indicators, and the list of what it is escalating with the reason.
The connective tissue is one line at the top of the group’s minutes naming what is going up and why, and one line in the board’s minutes recording what came back down. Without those, the two bodies communicate by rumour and the third failure mode is inevitable.
The right reflex. Before an item is discussed, ask which body can take the decision it will produce. If the answer is « this one », proceed. If it is « the other one », stop and escalate with the reason written down. If it is « neither », the boundary itself is the defect, and fixing the boundary is a board item that will pay for itself several times over.
What an audit actually reads
Inspectors do not read minutes for the discussion. They follow a thread: a report enters the system, becomes a hazard, is assessed with a severity and a likelihood, produces a mitigation, produces an action with an owner and a date, and produces evidence that somebody checked whether the mitigation worked. Any break in that chain is the finding, and two breaks recur more than the rest — a residual risk accepted by somebody without the authority to accept it, and an action closed with no evidence beyond the fact that its due date passed.
The second thing an audit reads is attendance. A safety review board chaired by a delegate, meeting after meeting, is evidence that the accountable manager’s accountability is nominal, and no amount of documentation repairs that impression. It is also, in practice, the reason boards drift into reviewing occurrences: a chair without the authority to commit resources can only discuss.
Small organisations, and the cadence question
An organisation of thirty people should not run two committees, and nothing in the rule says it must. What it must do is show, in one set of minutes, which decisions were taken in which capacity — that the residual risk on a given item was accepted by the accountable manager, not by the meeting as an atmosphere. A single meeting with a two-part agenda and a clear record of who accepted what is entirely defensible. A single meeting that blurs the two is the small-organisation version of the second failure mode.

Cadence is a matter of practice rather than regulation, and the honest guidance is that it should follow the rate at which your data changes. A group that meets monthly when the reporting rate produces a handful of items a quarter will manufacture business to fill the slot; a board that meets annually cannot respond to an indicator that went through its alert level in March. Set the intervals from the data, review them once a year, and treat a meeting that has stopped producing decisions as a symptom rather than a scheduling problem. Safety managers comparing structures across operators — the sort of exchange that fills the governance threads on the FL330 forums — converge on the same conclusion: the organisations that escaped the recurring agenda did it by writing down a ceiling, not by redesigning the meeting.

Leave a Reply
You must be logged in to post a comment.