Most Workplace Accidents Are Preventable. So Why Do They Keep Happening?

[S_adna Smallwoo]

Most Workplace Accidents Are Preventable. So Why Do They Keep Happening?

Every safety manager has lived through this moment. An incident occurs. It gets investigated, closed, and filed. Three months later, a similar incident happens. Then another. Only in retrospect does someone notice the thread: same shift, same location, same category of task. Three incidents that, together, pointed to a systemic risk, but were only connected after the harm had already repeated.

This is the core challenge the ServiceNow Health and Safety Incident Pattern Detector is built to solve.

The Problem With Manual Pattern Recognition

Safety investigators are skilled professionals. But the volume and complexity of incident data across a large organisation makes manual pattern detection genuinely hard. Spotting recurring themes across dozens or hundreds of records, correlating location clusters with time-of-day trends, identifying whether the same contractor team or asset category keeps appearing at the centre of incidents. This takes time that most safety teams simply don't have.

The problem is compounded when incident data lives in silos. When similar near-misses occur across different sites, teams, or reporting periods, the patterns remain hidden because the data never connects. Without unified methodology, lessons learned from one incident rarely transfer to the next, meaning identical hazards resurface under different job numbers.

The result is that preventive action lags the problem. By the time a pattern is identified, it may already have caused repeat harm, regulatory exposure, or significant cost. The National Safety Council puts the average workplace injury at over $42,000 in direct costs alone, not counting lost productivity, rehiring, insurance increases, and equipment downtime.

What the Incident Pattern Detector Does

The Incident Pattern Detector is an AI-powered capability built natively into ServiceNow Health & Safety. It automatically identifies recurring patterns across H&S incidents by scoring similarity across multiple dimensions simultaneously:

• Location — are incidents clustering around a specific site, zone, or facility?
• Time — do incidents concentrate in particular shifts, days of the week, or hour blocks?
• People — are the same teams, roles, or individuals repeatedly involved?
• Assets — is a particular piece of equipment or asset category consistently present?
• Injury type — are the same categories of physical harm appearing repeatedly?
• Severity — is the seriousness of harm consistent across the cluster?
• Incident category — is there a recurring type of event beneath the surface?

Rather than treating these dimensions in isolation, the system combines them into a composite similarity score and classifies each detected pattern into one of three confidence tiers:

Tier	Score	Meaning
Low	< 40	An emerging signal worth watching — not yet statistically robust, but warranting attention
Medium	40–69	A meaningful pattern supported by sufficient data — grounds for investigation and intervention
High	≥ 70	A high-confidence, recurrent pattern — requires immediate preventive action

This tiering reflects the academic consensus on what makes pattern detection actionable: it's not enough to flag a cluster; you need a confidence signal that tells the investigator how much weight to give it.

What Makes This Different

A few things distinguish this capability from simpler reporting or analytics approaches:

It's multi-dimensional. Most safety dashboards let you filter by one or two dimensions. Pattern detection operates across all of them simultaneously, finding clusters that no single filter would surface. This is exactly the kind of complex, multi-factor correlation that the academic literature identifies as the primary source of missed patterns.

It's scored and tiered, not binary. Rather than simply flagging "pattern found / not found", the system gives investigators a confidence signal they can act on proportionately.

It connects data that would otherwise stay siloed. The system operates across your full incident history on a single platform, eliminating the data fragmentation that the research consistently identifies as the primary reason patterns go undetected across teams and sites.

Security-Aware by Design

One of the harder design problems in incident pattern detection is what to do with sensitive records: incidents involving sensitive personal data, under legal hold, or flagged for restricted visibility.

The Incident Pattern Detector handles this carefully: sensitive incidents are included in the statistical analysis (so the pattern detection remains accurate), but their details are masked for users who lack the appropriate role. The pattern score and tier reflect the full dataset; the drill-down is governed by RBAC. This ensures that safety insights aren't degraded by data access limitations, while preserving the privacy and access controls that sensitive incidents require.

Scheduled Detection to fit Your Workflow

The feature is designed to fit into real safety workflows, not replace them. Detection can be triggered by Scheduled detection where a background job runs on a configurable cadence, continuously preprocessing incident data and surfacing patterns as they emerge. The system uses automated change detection, tracking shifts in incident membership and data timestamps, to ensure only relevant records are reprocessed on each run.

You can then access the identified patterns in the Pattern analysis dashboard in the Health and Safety Workspace. This is the single pane of glass you will use to manage and mitigate incident patterns.

Natural Language Summaries and Preventive Action Plans

Surfacing a cluster of similar incidents is only half the job. The capability integrates with Now Assist to generate human-readable summaries of each detected pattern, written in plain language with links to the relevant incident records and a breakdown of which dimensions drove the match.

Beyond summaries, the system proposes corrective and preventive actions tailored to the specific pattern. These aren't generic recommendations; they're grounded in the pattern's dimensions and generated through Now Assist's reasoning capabilities, giving investigators a concrete starting point rather than a blank page.

This matters because the research consistently shows that the bottleneck in safety improvement is not investigation capacity. It's the time between identifying a pattern and taking structured preventive action. Shortening that gap is where the measurable reductions in incident rates come from.

Looking Ahead

The Incident Pattern Detector is the foundation for a broader direction in AI-native safety management. Given that 70% of human-error-related accidents are preventable through management action, the question for safety leaders is no longer whether the data contains the signals needed to act earlier. It almost certainly does. The question is whether you have the tools to find them before the next incident closes another record that will be investigated in isolation, filed, and forgotten.

The H&S Incident Pattern Detector is part of the ServiceNow Health and Safety portfolio, available as part of the Now Assist for Health and Safety. Check out my colleague Claire's blog post detailing the March 2026 Health and Safety release for more info.

[shahbasshhbs]

Research shows that a significant proportion of incidents are linked to human behavior, often influenced by factors such as time pressure, Peer pressure, Safety awareness, Attitude, fatigue, and cognitive bias rather than simple negligence.Workers

don't intentionally take risks; they make decisions within the constraints of real work environments.

One key reason accidents persist is the gap between "work as imagined" and "work as actually performed." Procedures may look perfect on paper, but real-life conditions-deadlines, resource limitations, and operational pressures-drive people to adapt, sometimes unsafely.

Another issue is the normalization of deviance. When unsafe practices don't immediately result in harm, they gradually become accepted as "the way we do things," increasing long-term risk.

Additionally, many organizations remain reactive rather than proactive-learning only after incidents occur instead of addressing weak signals and near-misses early.

I can recall an incident caused by one of our Subcontractor by entering into the sea water tank during off work hours which was identified as a confined space; Neither the worker was authorized to enter confined space nor the activity was documented. No Permit to work procedures were followed and Confined space entry requirements were fulfilled. The task supervisor from the sub-contractor forced the worker to take shortcut and made him to enter the tank which resulted in a fall Incident from a height of 4 meter ended up with a compound fracture and the worker undergone for a surgery.

This incident were easily avoidable as there wasn't any requirement for the worker to enter the sea water but the cavalier attitude and peer pressure triggered him to commit such a blunder.