How airlines detect emerging risk
before incidents occur.

See the highlighted block above. This is the prescriptive companion to why mature SMS programs still miss emerging risk: that article diagnoses the blind spots; this one sets out what to do about them.

It is worth being precise, because the language around this is often overheated. Detecting emerging risk does not mean predicting the next incident. No system tells an airline the date, the flight or the failure. Selling that certainty is the fastest way to lose a safety team's trust.

What detection means is earlier sight. Serious events are almost always preceded by conditions that were, in hindsight, observable: a barrier weakening, precursors recurring, signals converging, an operation drifting. Detection is the discipline of seeing those conditions while they are forming, early enough that acting is cheap and undramatic — a quiet adjustment rather than an investigation. The honest claim an airline can stand behind is “we saw it forming and acted,” not “we predicted it.” That distinction is not modesty; it is what keeps the practice credible and the team honest.

Earlier sight comes from a small number of practices, each aimed at a different blind spot. Used together they cover the ways emerging risk hides.

1. Cross-module correlation. The single highest-value practice. Most emerging risk is a combination that no one module can see: a weakening barrier, an expiring competency and a cluster of minor occurrences that all touch the same hazard. Reading those together — routinely, not only during an investigation — is what turns three unremarkable records into one visible risk. This is the practice that directly attacks the silo blind spot.

2. Precursor and weak-signal tracking. For the hazards that matter most, identify what reliably precedes them and watch it deliberately. Aggregate weak signals against the structures they touch and escalate on convergence, not on raw volume — so a small number of well-evidenced “worth a look” items reach a human, rather than a flood of flags.

3. Barrier-health monitoring. Treat the barriers in your bowtieas living things with a current state — effective, weak, failed — updated as they are tested and as events touch them. A barrier quietly degrading across several minor events is one of the clearest early pictures of risk there is, and it is invisible if barriers are a one-time diagram in a file.

4. Drift checks. Look for operational drift on purpose, because it never reports itself. Compare work-as-done with work-as-imagined through normal-operations observation and line monitoring, and watch the slope of adherence trends rather than waiting for a threshold breach.

5. Balanced leading indicators. Carry a focused set of leading indicatorswith a real causal story, each paired with the lagging outcome it claims to anticipate. Read them for direction and trend — the approach to a limit, not only its breach.

Practices only detect risk if they run on a cadence. Detection is a rhythm, not a project, and the rhythm is what an executive actually owns. A workable shape:

• Continuous — intake and triage. Reports, data events and findings are captured and classified consistently as they arrive, against a stable taxonomy. Consistency here is what makes every later pattern trustworthy.
• Weekly — the cross-read. A standing review that reads across modules for the week: which barriers moved, which signals converged, which precursors recurred. The output is a short list of forming risks worth attention, each carrying its evidence.
• Monthly — the board question.The safety review board spends its time on “what is forming and what are we doing about it,” not only “what happened and what did we close.” The agenda is the lever: it decides whether the organisation looks forward or back.
• Quarterly — the deep look. A periodic examination of drift, indicator validity and the hazards most worth instrumenting next. This is where leading indicators are checked against their lagging outcomes and pruned if they have stopped meaning anything.

The exact intervals matter less than the principle: forming risk is reviewed on a schedule that is faster than the speed at which it matures into events. A quarterly trend review alone is too slow, because it samples late and misses what assembled in between.

The practices and the rhythm depend on a few conditions. Without them, the effort produces motion without sight.

• Connected data.The non-negotiable foundation. If the modules cannot be read together, cross-module correlation — the highest -value practice — is impossible. This is the Connected level of the maturity model, and most missed risk traces back to its absence.
• A consistent taxonomy. Patterns across records are only trustworthy if the records mean the same thing. Inconsistent classification quietly invalidates every trend built on top of it.
• A Just Culture.Detection runs on the faint, the uncertain and the unflattering — the near-miss filed honestly, the drift named openly. A Just Culture is what keeps that information flowing; without it, the signals dry up and there is nothing to detect.
• Human judgement in the loop.Tools and analysis propose; qualified people decide. The value of detection is putting a small number of credible, evidenced patterns in front of the right person early — the judgement about what they mean and what to do stays human. The controls that make any AI assistance defensible are covered in AI in aviation safety management.

Assembled, these practices and conditions are the operational definition of aviation safety intelligencein action. The point is not the sophistication of any single technique; it is an operation that can see itself whole and look forward on a rhythm. That is what lets an airline act on risk while it is still forming — quietly, early, and before it becomes an incident.