One incident. Five obligations. Different forms, different clocks, all running at once.
Last updated
Zepth Core module
8 hrs
to report a workplace fatality to OSHA — 24 for a hospitalisation, amputation or eye loss
OSHA 1904.39
20–91%
the range across the research for how many workplace injuries go unreported
Systematic review
The width of the range is itself the finding: nobody knows how much is missing, which is the problem.
4.5×
indirect cost against direct cost for smaller injuries — and almost all of it uninsured
OSHA Safety Pays
2.2
recent US construction TRIR — the benchmark your own number gets read against
US Bureau of Labor Statistics
An incident is anything from a near miss to a fatality — and the difference between a learning organisation and a liability is what happens in the hours and days afterwards.
Four things decide which one you are: classification, the reporting clocks, the quality of the investigation, and whether the metrics are honest. Get them right and incidents get rarer. Get them wrong and the same failure returns at higher energy.
The reporting clocks are unforgiving, and they run in parallel. In the US, a workplace fatality must be reported to OSHA within 8 hours; an in-patient hospitalisation, amputation or eye loss within 24. The UK’s RIDDOR requires deaths and specified injuries reported without delay, and over-seven-day incapacitations within 15 days of the accident. In the UAE, work injuries must be notified within 24 hours, though the reporting channel varies by emirate. Saudi Arabia runs a 24-hour accident notification alongside a three-day insurance window.
And on most GCC and enterprise projects the contract adds a second clock, often a faster one: immediate or 24-hour notice to the engineer or client, a draft investigation in 48 to 72 hours, a final report with corrective actions in 7 to 14 days. Missing the contractual clock can trigger a default notice even when the statutory report went in on time.
One incident, five obligations, five different forms, all of them running at once while the site is still in shock. This multi-regulator fan-out is precisely where manual processes fail — and it is the part software genuinely solves.
OSHA’s own cost model puts indirect costs at roughly 4.5 times direct for smaller injuries — crew standdown, supervision time, retraining, replacement labour — and nearly all of it is uninsured, on contracts running 3–5% margins. The multiplier falls as the injury gets larger, which is the opposite of most people’s intuition: it is the small injury whose hidden tail is disproportionate.
Then the incident prices your future. In the US, injury frequency drives the experience modification rate that scales your premiums and gates your bidding — above roughly 1.0 to 1.2 commonly excludes you from prequalification outright. In the GCC, client HSE questionnaires demanding three-to-five-year LTIFR and fatality history perform exactly the same gating by a different instrument. The tail can outcost the incident.
Classification discipline drives everything downstream. The severity ladder — near miss, first aid, medical treatment, restricted work, lost time, fatality, plus “dangerous occurrence” for the no-injury high-potential event — determines what gets reported to whom, what gets investigated how deeply, and what enters TRIR versus LTIFR. One convention trips up multinationals constantly: TRIR uses a 200,000-hour base and LTIFR a 1,000,000-hour base. Comparing them across regions without converting produces nonsense, and then someone acts on it.
Honest metrics, or none. Every practitioner knows the games. A “suitable duties” programme that returns an injured worker to light duty on day one makes the lost-time injury vanish from the numbers without changing anything about the injury. That is why sophisticated clients now demand TRIR, DART and SIF-potential rates together — no single number survives contact with a determined case manager.
Near-miss culture is the leading indicator. Underreporting is the norm rather than the exception: systematic reviews put the share of workplace injuries that go unreported anywhere from 20% to 91%, and industry estimates suggest the large majority of near misses are never reported at all. Every metric that punishes reporting — TRIR-linked bonuses, “days since last incident” boards — suppresses exactly the data that predicts serious events. Rising near-miss counts with falling severity is what health looks like, and it looks like failure on a dashboard.
Match the investigation to the potential, not the outcome. 5 Whys for low-severity events: fast, but linear and heavily investigator-dependent. ICAM or a comparable systemic method — failed defences, individual actions, task conditions, organisational factors — for anything with serious potential, because it forces organisational factors into view instead of stopping at the operator. Universal practice either way: preserve the scene, photograph before anything is disturbed, quarantine the equipment, and take witness statements within 24 to 48 hours — separately, because witnesses left together converge on a shared narrative and you lose the independent accounts.
The famous safety triangle holds that a fixed ratio of minor incidents underlies each major one, so driving down the minor ones drives down the major ones. The modern critique — and it matters — is that this is only half true.
Minor-incident data predicts minor incidents. Fatalities and serious injuries have their own distinct precursors, and they are about energy: work at height, lifting operations, energised systems, mobile plant. A site can drive its cut-finger rate to zero and still kill someone, because nothing it did addressed the crane.
Which is why fatality prevention requires tracking high-energy precursors separately, and why a near miss with serious-injury potential deserves a full investigation at zero injury. The outcome was luck. The exposure was real.
An unreported near miss leaves the hazard in place, and the same failure mode returns at higher energy as a recordable. The late report earns a citation and, worse, a presumption of poor management that colours the whole follow-up inspection. The blame-first investigation produces a name rather than a cause, so the cause repeats. And the paper log means the contractual 48-hour draft deadline arrives before anyone has assembled the timeline.
Every one of those is a records-and-workflow failure before it is a safety one. That is the whole argument for running incidents on a system rather than in a binder.
One incident record drives everything. Classification sets the reporting obligations and starts their clocks. Notifications route to the right parties with deadlines visible. Investigation evidence — photographs, statements, the timeline — attaches to the record rather than living in someone’s phone. Corrective actions carry owners, deadlines and verified closure.
TRIR, LTIFR and DART compute from live data with both hour bases handled, so a regional comparison is a conversion rather than a mistake. The multi-regulator fan-out becomes a checklist with timestamps instead of a scramble.
Five clocks run themselves, so a contractual deadline cannot be missed while the statutory one is being met.
The investigation is matched to what could have happened, not to what did — which is the only way SIF exposure ever gets found.
The metrics resist gaming, because TRIR, DART and SIF-potential are trended together rather than one at a time.
The near-miss data survives, because nothing in the system punishes the person who reports.
The severity classification determines the reporting duties and starts their clocks — statutory and contractual, in parallel.
OSHA, RIDDOR, UAE, Saudi and the contract, each with its own deadline and time remaining. The fan-out becomes a checklist.
Photographs, witness statements, timeline and equipment status on one record — with 5 Whys or ICAM structure to match the potential.
Owners, deadlines and evidence. A root cause with no assigned action is a report, not a fix.
Computed from live data on the correct hour base, so a 200,000-hour TRIR is never compared naively against a 1,000,000-hour LTIFR.
High-energy precursors — height, lifting, energised systems, mobile plant — tracked separately from minor-incident counts, because they do not predict each other.
From the field, with evidence. Classification is the first decision because it sets everything after it: who must be told, how fast, how deeply it gets investigated, and which metric it lands in.
Every statutory and contractual deadline the classification triggers, on the board with time remaining. The 8-hour fatality clock does not wait for someone to look it up.
Scene preserved, photographs before disturbance, equipment quarantined, witness statements inside 24–48 hours and taken separately. Method matched to potential severity: 5 Whys, or ICAM.
Corrective actions with owners, deadlines and evidence-based closure — not a report that ends at a root cause nobody was assigned.
TRIR, LTIFR, DART and SIF-potential from live data, on the right hour base, trended together — so no single number can be managed in isolation.
From the moment of classification, every statutory and contractual deadline is on the board with time remaining. The 8-hour fatality clock does not wait for someone to check a manual — and the contractual clock, which is often the faster one, does not get forgotten because the statutory one was met.
Investigation reports and client notifications drafted from the incident record — timeline, evidence, personnel — for the HSE manager to verify and sign. The 48-hour contractual draft stops being a scramble.
Near misses and violations correlated against incident history, so the SIF-potential patterns — height, lifting, energy, mobile plant — surface before they convert into the thing they were rehearsing.
Divergence between the lost-time trend and the total-case trend gets flagged. That gap is the signature of case management rather than safety improvement — and it is invisible if you only watch one number.
The engineer’s judgment stays in charge; the AI removes the latency and the blind spots.
Incident dashboards with classification, corrective-action ageing and trend analysis — and TRIR, LTIFR, DART and SIF-potential computed from live data on the correct hour bases. The lost-time trend and the total-case trend are shown together, because a divergence between them is the thing worth seeing. Exportable for statutory and client safety reporting.
Severity classification. A near miss caused no injury but could have. A first-aid case needs only basic treatment and is not recordable in most regimes. A recordable involves medical treatment beyond first aid, restricted work, lost time, or worse. The classification determines your reporting duties and which metric the event lands in.
US: fatalities within 8 hours, hospitalisations, amputations and eye loss within 24. UK: without delay for deaths and specified injuries. UAE: 24 hours, though the channel varies by emirate. Saudi Arabia: 24 hours, plus a three-day insurance window. And your contract almost certainly adds its own clock to the client — often a faster one, and missing it can trigger a default notice even when the statutory report was on time.
TRIR = recordable cases × 200,000 ÷ hours worked. LTIFR = lost-time injuries × 1,000,000 ÷ hours worked. The bases are different, so convert before comparing — a TRIR and an LTIFR are measured on different rulers. Recent US construction TRIR sits around 2.2, which is the benchmark your own number gets read against.
Read the full answerMatch the method to the potential severity, not the actual outcome. 5 Whys for simple, low-severity events — it is fast, but linear and investigator-dependent. ICAM or a comparable systemic method for anything with serious potential, because it forces organisational factors into view instead of stopping at the operator.
Directionally, yes — but the modern evidence rejects the fixed ratios of Heinrich’s triangle. Minor-incident data predicts minor incidents. Serious injuries and fatalities have their own high-energy precursors: work at height, lifting, energised systems, mobile plant. Track and investigate SIF-potential events specifically. Counting cut fingers will not save anyone from the crane.
Read the full answerOSHA’s model puts indirect costs at roughly 4.5 times direct for smaller injuries, falling proportionally as the injury gets larger — and almost all of it uninsured. Then comes the tail: the experience-modification and premium impact, and prequalification exclusion. The tail can outcost the incident.
Read the full answerRelated modules
Related answers
Terms defined here
Zepth is the construction project delivery platform — it runs construction, procurement and asset management on one record, and does the work: reading the drawings, reviewing the submittals, matching the invoices and flagging the risks, with a human sign-off on anything consequential.
A short, tailored walkthrough on your real workflow — no generic demo.