A line-opening file arrives at the safety committee with a risk assessment concluded as « tolerable subject to reservations », with no one actually holding the reservations and the date already announced to the crews. The safety manager will have to say, that day, that the sentence is worthless. An interview that never checked whether they are capable of it has measured something else entirely: a matrix, a taxonomy, a dashboard. Useful, never decisive.
What the operator is really buying with this position
Analytical competence is available on the market. Building a hazard map, maintaining an event taxonomy, producing indicators that stay consistent from one quarter to the next: this can be learned and verified on paper. What the operator actually pays for is the ability to hold an unfavourable conclusion in front of the person who signs the postholder’s annual appraisal.
The safety management system does not produce safety: it produces decision-making information. Complacent information is not degraded, it is worthless: it gives management the feeling of oversight where none remains. A dashboard that has stayed green for two years describes a system that has stopped looking.
Hence the asymmetry of the position. Other functions are assessed on what they enable: rotations, an entry into service, an approval. This one is assessed on what it prevents, and no one knows how to measure an event that never happened. Its normal output consists of creating cost, and its independence is not shown at interview: it is reconstructed from dates, wording and signatures.
A safety manager’s real job isn’t just making reports or charts. It’s about having the courage to say “no” when something is unsafe, even if it delays plans. The best safety managers make decisions that might cost money or time, but keep people safe.
At a regional airline, a safety manager noticed a risk assessment for a new route had been marked “tolerable” without clear follow-up. They insisted on adding concrete mitigation measures before approval, delaying the route launch by two weeks. The airline later avoided a serious incident due to those measures.
Independence is not a character trait, it is a construction
There is a temptation to look for a temperament in this role. That is an analytical error. A determined postholder, placed in a construction that subordinates them to the activity they assess, will lose quickly; a measured postholder, in a clean construction, will hold without raising their voice. Five levers decide this, and they can be verified on paper: the line of reporting, effective access to the accountable manager, dedicated resources, the hand that signs off acceptance of residual risk, and how the organisation handles combining the role with another activity the postholder also assesses.

The European provisions applicable to operators, and the accompanying means of compliance, require that the person in charge of safety management be able to address the accountable manager directly. A text can require that the route exist; it cannot require that it be used. In some organisations it has never been taken, and nothing in the manual says so.
A good candidate questions this construction before talking about themselves. The table sets out, for each lever, the wording that reassures and the fact that proves.
| Lever | Apparent construction | Effective construction | Evidence to request |
|---|---|---|---|
| Reporting line | Reports to senior management on the org chart | No superior in production sets their objectives | Who signs their annual appraisal |
| Access to the accountable manager | Direct access written into the manual | At least one referral outside committee during the year | Log of referrals, dates and outcomes |
| Resources | Budget pooled with another support function | External expertise that can be mobilised without prior arbitration | The last expert study commissioned, and who authorised it |
| Risk acceptance | The safety manager signs the form | Signed by the one who decides, not the one who analyses | The last three residual-risk acceptance forms and their signatory |
| Combined roles | No combined role declared | Combined role declared, with a written recusal rule | A case where the recusal was applied |
The questions that get people talking, and those that serve no purpose
« Tell me about a disagreement with your management » measures nothing. Everyone has a story of a disagreement won, told with the comfort of hindsight. The account is sincere and perfectly useless: it describes the easy case, the one where the organisation eventually gave way.
The questions that matter demand the opposite: a recommendation that was not acted on, a note with a known date. They target the written record, not opinion: what was written, to whom, in what form, and what the recipient did with it. Someone who has actually held the role answers with objects; the others answer with adjectives.
| Question | Weak answer | Strong answer | What it measures |
|---|---|---|---|
| A recommendation of yours that was not followed? | « Never, I’ve always been listened to » | A dated case, the decision made, the reason given | Whether the role has ever been put under tension |
| What survives in writing? | « It was discussed at a meeting » | An entry in the minutes, with a review deadline | The ability to make a decision reversible |
| A risk downgraded? | « The initial rating was too severe » | What had actually changed before the downgrade | The difference between a barrier and an argument |
| A report implicating a manager? | « I would have treated it like any other » | The route followed, who was excluded, what the reporter received back | The real solidity of the just culture |
The report implicating an in-house manager
The case is common: an event report points, without naming names, to a practice put in place by a manager currently in post. Protection of the source only counts if it holds there, because that is where a reporter observes whether the system protects or exposes. The expected answer is a route — who receives it, who is excluded because of involvement, within what deadline the reporter gets feedback, and what it contains.
The trap: taking the absence of conflict as proof of influence. A candidate who claims to have always been listened to describes two possible situations, and neither serves them: either they never made a costly recommendation, or the organisation never put them under tension. You then have no data on their behaviour on the day the conclusion clashes with a date already sold.
The lost disagreement and the record that must remain
The decisive point comes down to one sentence: the person who analyses the risk is not the one who accepts it. The postholder assesses, proposes barriers, gives an opinion; acceptance of residual risk belongs to the accountable manager or an explicitly designated delegate. A manager who signs the form themselves has traded their independence for a signature, and can no longer contest a decision they co-signed.
When the recommendation is not followed, good practice is neither resignation nor silence: it is the entry in the record. It takes the form of a neutral note carrying four elements: the recommendation as it was formulated, the decision taken, the reason given, the review deadline and its trigger criterion. This is not a defensive act: it is what makes the decision reversible, whereas an undocumented decision will simply be defended.
The right reflex: ask the candidate to write, in front of you, in five lines and ten minutes, the minute-book entry for a recommendation that was not followed, based on a case you give them. The exercise reveals what the spoken answer hides: neutrality of tone, precision of the deadline, presence of a review criterion, the ability to write down a disagreement without putting a person on trial. Keep the text — it compares better than an impression.
The band where independence is really tested
A risk matrix is decisive in only two corners. Catastrophic and frequent: no one argues. Minor and extremely rare: no one argues either. Between the two runs a band where two competent professionals, with the same data, classify differently without either being wrong. This is the domain of judgement, and the only place where the pressure is applied.
A risk shifts from one cell to another in three ways: probability changes because new data has come in, severity changes because a consequence scenario has been revised, or the barrier changes because a real measure is in place. Only the third is a safety action; the first two remain legitimate only if something has changed in reality and not just in the calendar.
Hence a simple question: a case where the candidate downgraded a risk, and what had changed at the moment of the downgrade. A barrier installed, new data, external operational experience feedback: you have a professional. A discussion, a consensus, a rating judged too severe after the fact: you also have an answer.
When deciding if a risk is acceptable, the hardest calls are in the middle ground. If someone changes their mind about a risk, it should be because something real changed—not just because they had a discussion. A good safety manager can explain exactly what changed to make the risk lower.
At an airline operating Airbus A320s, a safety manager downgraded a risk related to runway overruns. The change was justified by new data showing improved braking performance on a recently resurfaced runway (Runway 27L at Paris Charles de Gaulle, 3,600m long, with new friction coefficients measured at 0.65). The downgrade was based on real evidence, not just opinions.
Combined roles, a central issue for smaller organisations
At a smaller operator, the postholder frequently combines roles: chief pilot, head of training, compliance monitoring manager, sometimes several of these at once. Combining roles is not abnormal — it is accepted when the size and complexity of the organisation justify it and the authority accepts it — but it installs a conflict that no goodwill dissolves: one assesses the risk of a set-up one has authored.
Combining safety management with compliance monitoring is the most deceptive pairing: the two functions look neighbouring and look at different things. Compliance measures the gap to the letter of the text; safety measures the gap to reality, even when the text is respected. Confusing them means giving up the second pair of eyes on the very day the first is satisfied.
The strong answer does not deny the conflict: it names it and describes a mechanism. A written recusal on files touching one’s own activity, analysis entrusted to a peer designated in advance, a cross-review with a comparable operator, a note in the file whenever the recusal is applied. The weak answer fits in one sentence heard often: « I know how to separate the two ».
The simulation exercise, because behaviour cannot be declared
A type rating is not issued on the strength of a questionnaire: the candidate is placed under load, the situation is loaded, and what gives is observed. There is no reason to assess a role whose value lies in behaviour under pressure any differently.
The protocol is cheap: a real anonymised file, an uncomfortable conclusion, twenty minutes, and a panel member playing the decision-maker who has already committed to the date. What is observed is not the conclusion, written in advance, but the path taken when it is refused.
Three signals read without ambiguity. Lexical drift: from « unacceptable as it stands » to « acceptable under enhanced vigilance » without any barrier having been added. The phantom barrier: a mitigation measure with no owner, no deadline and no way of verifying it is in place. The deferred review: a review date set after the period during which the feared event could occur.
The favourable signal is symmetric: an engineering compromise rather than a vocabulary compromise. A temporary restriction, a slot limitation, a reinforced crew on the first rotations, raised minima while data accumulates. The candidate gives ground on the calendar, never on the conclusion.
The same exercise tests the sense of timing: ask the candidate at what point in the file they should have written their first reservation, and why. An objection raised at the committee that finalises the decision costs far more than a reservation lodged three weeks earlier, when the information was already sufficient.
What to check in your own organisation before meeting the first candidate
Three checks take half a day and change the interview. Count, over the last two years, the recommendations formally not followed and recorded: zero is not a good result, it is a warning. Look at who signed the last residual-risk acceptance forms. Check whether the safety committee has ever met without the accountable manager: chaired permanently by delegation, it is no longer a management body.
One requirement remains, absent from every text, yet it decides the course of the first technical disagreement: operational credibility. A postholder who has never held the function they are contradicting will find themselves dismissed for their ignorance as soon as the discussion drops a level: the safety debate becomes a legitimacy debate. Good profiles come from varied backgrounds, as shown by the exchanges between postholders on the FL330 forum; what remains is to check how each will make themselves technically credible.
A grey area remains that recruitment cannot close: independence cannot be written into a contract. What can be defined from the letter of engagement, however, is an observable event at six months — the first recommendation not followed and recorded, the first direct referral to the accountable manager — and made a shared review point, never a numeric target, which would manufacture exactly the facade one is trying to avoid.
The applicable regulations and the operator’s approved documentation, including the job descriptions accepted by the authority, prevail in all circumstances over the points of method given here.
Questions fréquentes
Can the safety manager sign off acceptance of a residual risk themselves?
That is not their role. They assess, propose barriers and give an opinion; acceptance of residual risk belongs to the accountable manager or an explicitly designated delegate in the approved documentation. A postholder who co-signs the acceptance can no longer contest a decision they were party to. Checking who signed the last acceptance forms is the fastest test of the system’s real independence.
Which question best distinguishes a good candidate at interview?
The one about a disagreement that was lost, never one that was won. Ask for a recommendation that was not followed, its date, the reason given by the decision-maker and what remains of it in writing. A candidate who has genuinely held the role answers with objects — minutes, a form, a review deadline — not with adjectives. A career with no recommendation ever turned down signals a role that has never been put under tension.
Does combining the role with an operational function disqualify a candidate?
No, and it is common among smaller operators, where it is accepted when the size and complexity of the organisation justify it and the authority accepts it on file. What matters is how the resulting conflict is handled: a written recusal rule on files touching one’s own activity, analysis entrusted to a peer designated in advance, systematic notation in the file. A candidate who only answers that they know how to separate the two has never had to apply that mechanism.
How can behaviour under pressure be tested without an artificial role-play?
Give a real anonymised file with an uncomfortable conclusion, have a panel member play the decision-maker who has already committed to the date, and limit the exercise to about twenty minutes. What is observed is not the conclusion, known in advance, but the path taken: drift towards softer wording, a mitigation barrier announced with no owner or deadline, a review deadline set after the risk period. The favourable signal is an engineering compromise, not a vocabulary compromise.
What should you check in your own organisation before starting the recruitment?
Three things, in half a day. The number of recommendations formally not followed and recorded over the last two years, knowing that zero is a warning, not a good result. The signatory of the last residual-risk acceptance forms. And finally the accountable manager’s actual presence at the safety committee, since a committee permanently chaired by delegation is no longer a management body.
Can independence be guaranteed through the employment contract?
Independence cannot be written into a contract, because it plays out in situations the contract does not describe. What can be planned from the outset, however, is an observable event at six months — the first recommendation not followed and recorded, the first direct referral to the accountable manager — and made a shared review point. It must never become a numeric target: that would manufacture exactly the facade one is trying to avoid.

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