October 7, 2026
.jpg)
Across many industrial operations, overall accident rates have followed a steady downward trend for several years, reflecting the maturing of safety programmes and the consolidation of preventive routines in the field. In the European Union, for example, the number of non-fatal accidents at work has continued to fall, and yet Eurostat recorded 3,367 fatal accidents in 2024, 45 more than the previous year.
This discrepancy shows that a decline in common occurrences does not automatically translate into an equivalent reduction in fatalities. The gap between these two indicators exposes a critical weakness in conventional risk management models.
In day-to-day operations, frequency and severity are often treated as equivalent quantities, which produces a distorted reading of plant safety. While frequency counts the number of events that occurred, severity assesses the actual harm, or the destructive potential, that those deviations could have unleashed.
Lowering the overall count of minor incidents therefore guarantees no protection against the worst-case scenarios. It is precisely for this reason that independent analysts such as Verdantix and Gartner place the prevention of serious injuries and fatalities (SIF) at the top of the industrial sector's strategic priorities.
Much of the classic prevention culture has its origin in the pyramid formulated by Herbert Heinrich in 1931. Drawing on the empirical analysis of thousands of accidents, he proposed a fixed mathematical ratio between no-injury occurrences, minor injuries and fatal accidents, arranged in a hierarchy in which a broad base supported a narrow top.
The visual clarity of that model made it easier to formalise the first industrial safety programmes, in an era that still lacked systematic records. The analytical distortion, however, came from the direct deduction that took hold in the following decades.
The premise emerged that reducing minor incidents at the base would almost automatically reduce the frequency of fatalities at the top of the pyramid. That interpretation led generations of managers to chase frequency targets, assuming that fewer minor injuries would inevitably mean a shop floor immune to tragedy.
The flaw in this assumption comes from the fact that frequency and severity respond to entirely distinct causes and dynamics. A superficial cut to a finger and a fall from height are not the same event in different degrees of luck, because they have completely different origins, propagation mechanisms and safeguards.
Concentrating effort on eliminating the first deviation, because it is visible and recurrent, creates a false sense of control that leaves untouched the conditions feeding the second. Statistical progress on the dashboard ends up concealing the operation's vulnerability.
Research by the Campbell Institute, linked to the United States' National Safety Council, shows that most low-severity injuries do not share root causes with serious and fatal occurrences. The analytical obstacle persists when organisations judge each event by its immediate outcome rather than assessing its potential destructive energy.
Under this skewed criterion, an extreme-risk operational anomaly that ends without casualties is filed away as a minor data point. That practice ignores what is, in reality, the rehearsal of a catastrophe.
Correcting this limited reading requires adopting the concept of a SIF precursor, which shifts the focus from the verified consequence to the plausible harm of the occurrence. A precursor is any technical condition or unexpected deviation with the potential to cause serious injury or death, regardless of whether any human harm occurred at the time.
A suspended load swinging beside a workstation, or an open pipe without confirmed depressurisation, are not negligible incidents. They are critical events whose severity simply did not materialise because of chance variables.
Adopting this reading inverts the way field teams respond to daily records. Instead of waiting for a failure to cause injury before opening an investigation, each event is classified straight away by its severity potential.
At this point, the architecture of a modern EHSQ (Environment, Health, Safety & Quality) software proves indispensable. With dynamic forms that guide the operator to characterise the scenario and identify the energy involved, the system ensures that the record is born already qualified and prioritised for immediate action.
Continuously classifying occurrences by potential solves the invisibility of risk, but it creates a new challenge of scale. As teams broaden their data capture to near misses and unsafe conditions, the volume of data grows quickly and makes it unfeasible to manually sift through thousands of reports to isolate the genuinely critical precursors.
The operational bottleneck thus shifts from a lack of information to an inability to process the excess of records. This is where artificial intelligence models act as a significant strategic lever.
Applied to open text descriptions, these algorithms group recurring operational patterns, identify the associated risk categories and highlight the lethal-hazard situations that would otherwise be diluted in the routine. The aim is not to replace the specialists' technical judgement, but to direct human intervention to where the vulnerability is real.
It is worth keeping a realistic stance, however, about the limits of these tools. Artificial intelligence depends entirely on the quality of the data collected, so histories with superficial descriptions produce weak diagnoses. Operational effectiveness therefore calls for an integrated approach that combines structured mobile forms on the front line with automated interpretation at scale.
Identifying precursors in time only delivers practical value if it leads to a deliberate reallocation of preventive barriers. Spreading management attention evenly across every deviation, treating any anomaly with the same weight, disperses the capacity to intervene and leaves the most dangerous activities exposed.
Technical priority should fall on rigid controls in the scenarios where the accumulated energy is enough to cause fatalities. The hierarchy of controls systematised by NIOSH remains the methodological reference for this approach, by favouring elimination and engineering protection before resorting to procedural rules or personal equipment.
Digital support makes this discipline feasible by ensuring that permits to work (PTW) and dynamic risk assessments are validated before critical tasks begin. In this way, the organisation makes certain that the containment barriers preventing serious accidents are active and confirmed, without relying on operators' memory or improvisation.
The entire analytical structure described loses its effectiveness if workers are afraid to report high-potential events. The continuous flow of data depends on the transparency of those working on the production line, and that communication stops the moment reporting comes to be seen as a risk of disciplinary punishment.
Approaches geared towards identifying individual culprits close investigations too early and leave untouched the organisational failures that James Reason classified as latent conditions of the system. A just culture establishes the opposite principle: it values the reporting of a near miss as an active contribution to collective protection, and it distinguishes unintentional error from deliberate violation.
That openness gives leadership back the vital signals it needs to strengthen operating procedures before any irreversible harm. Reducing the incidence of serious injuries and fatalities is, in the end, a strategic choice of focus and analytical rigour, one that moves away from the blind pursuit of statistical minimums in overall frequency.
Preventive balance comes from the ability to shift attention from injuries already sustained to the destructive energies that still lie latent in the field. By structuring processes at the source with integrated EHSQ software and by fostering an environment of operational transparency, companies trade the illusion of calm dashboards for a real safeguard of human life.
See how it helps you spot and act on high-potential signals before they become serious accidents
.webp){kind=small}
