The future of safety is prevention, not reporting
Module · Environment, Health and Safety
The metric almost every safety programme is steered by carries close to no signal about the events that injure, or worse.
Prophesee EHS replaces the count with a forecast, and the hope with proof.
EHS: a day in the life
One working day inside EHS, from the morning forecast to the evidence that the intervention worked.
Recordable injuries are at a twenty-year low. Fatalities are not.
That is not bad luck and it is not a reporting artefact. It is a known statistical result. The number almost every safety programme is steered by has been shown to carry close to no signal about the events that injure, or worse.
Sources: ILO, Apr 2025 · Hallowell et al, The Statistical Invalidity of TRIR, Professional Safety · BLS SOII Jan 2026 and CFOI Feb 2026 · EY Global EHS Maturity Study, Aug 2025 (n=526) · Benchmark Gensuite 2026 (n=260+) · IOGP Safety Performance Indicators, 2024 data.
From reporting incidents to preventing them
ForesightThe reporting system produces counts. There is no forward view by site and hazard, so the function reacts to last month.
Gradient-boosted trees tuned per metric and site, retrained monthly. Causal chains link leading indicators to incidents, and a counterfactual model flags which incidents a targeted intervention could have stopped.
HorizonInterventions are funded on argument. There is no model of the no-action trajectory, so no way to attribute a fall in incidents to an action.
Predict the no-action trajectory, set the target, model effectiveness, rollout and time to impact, then track actuals against plan. For the first time you can show which safety action moved the number.
PulseReporting rates vary ten to one between sites, so the near miss signal is confounded with reporting culture and nobody trusts it.
Rules run continuously across every site and contractor. Quarter-on-quarter trend up more than 10% in a country, or a spike inside a single week. Severity-scored alerts routed to a named owner.
NexusContractor identity and training sit in the contractor's own system under a different spelling, so employee and contractor risk cannot be scored on one scale.
How many incidents in Sweden last quarter, in plain language. Contractor compliance, training records and supplier self-assessments feed the same graph and the same risk score.
Turning safety challenges into decisions
The reporting system produces counts. There is no forward view by site and hazard, so the function reacts to last month.
ForesightContractor identity and training sit in the contractor's own system under a different spelling, so employee and contractor risk cannot be scored on one scale.
NexusReporting rates vary ten to one between sites, so the near miss signal is confounded with reporting culture and nobody trusts it.
Foresight PulseInterventions are funded on argument. There is no model of the no-action trajectory, so no way to attribute a fall in incidents to an action.
HorizonBoard questions are answered by assembling four systems by hand, so the answer arrives a week after the question.
Nexus12 AI applications that could be relevant
A sample of what becomes possible on the decision layer, not a fixed list: each application draws on the same data foundation and audit trail, and new ones are configured on the engines, not built from scratch.
Forward incident forecast by site and hazard, with the drivers behind each one.
Causal chains link leading indicators to incidents and open a prevention window.
A counterfactual model flags the incidents a targeted intervention could have stopped.
One risk score per site and contractor, recomputed as new data lands.
Write what bad looks like in plain English and have it watched continuously.
Contractor compliance, training and performance watched on the same rules as staff.
Model effectiveness, rollout and time to impact, and watch the projected curve update.
The date the current plan reaches the target, updated as actuals come in.
A project starting next month reshapes predicted risk before any work begins.
Ask how many incidents in Sweden last quarter and get the answer with sources.
Find every incident like this one, including the ones described in different words.
Supplier safety assessments completed online, read by the model, fed to the score.
Today: A monthly report of what already happened, assembled by hand from four systems.
ForesightSites ranked by predicted incident risk for the next 90 days, with the drivers behind each one.
Today: A contractor trend that nobody notices until the quarterly review.
PulseA contractor injury trend in Mexico breaches a rule she wrote in plain English. Scored, routed, acknowledged.
Today: Fund the intervention that argues loudest, then hope.
HorizonTwo interventions tested on the highest-risk site. The projected curve moves. She funds the one that shifts it.
Today: No way to show which action reduced anything.
HorizonLast quarter's intervention tracked against plan. Incidents fell as modelled. That goes to the board.
Tomorrow's risks become today's decisions.
Prevent the incident, not count it
We agree the metric and the baseline in week one, and measure the result on your data.