Same Hazard, Two Scores: Allocating Capital with a Risk Matrix
A 640-employee food-processing plant uses a 5x5 risk matrix to rank tasks for a $180,000 capital budget. Two maintenance crews independently assessed the same auger cleanout task: Crew A scored it 20 (severity 5 x likelihood 4); Crew B scored it 6 (severity 3 x likelihood 2). Six weeks ago an operator's sleeve was caught in that auger. The machine was stopped but not locked out, and the operator was pulling material out by hand. The plant has hired 47 new production workers in the past year, and turnover in the cleanout job is high. The safety committee wants to fund the Crew A score first and defer two other projects: a press-brake light curtain (score 12) and a forklift-pedestrian separation barrier (score 9) in an aisle with roughly 60 lift-truck trips per shift.
1. Why did two crews score the same task so differently, and what changes would make the scoring defensible enough to allocate capital?
Read the worked answer
Why the scores diverged
- The matrix is ordinal, so 20 versus 6 does not mean one hazard is 3.3 times worse. It usually means the criteria are undefined.
- Severity anchors are ambiguous: Crew B likely scored the typical outcome (a bruise or strain); Crew A scored the worst credible outcome (amputation or fatality). Both can be defended under a poorly written matrix.
- Likelihood without a stated exposure base (per task, per shift, per year) invites different assumptions.
- Individuals scored alone, with no calibration, no reference tasks, and no operator who actually performs the cleanout.
- Stored rotational energy, loose clothing, and a high-turnover crew were not captured as likelihood factors.
What to fix
- Write plant-specific anchors for each severity and likelihood level, with photographic examples and a defined time/exposure base.
- Score the worst credible outcome for fatality-potential screening, and keep that screen separate from the matrix number.
- Score in a team that includes the task expert; calibrate against three to five known reference tasks; document the rationale.
- Require re-scoring whenever staffing, material, tooling, or procedure changes.
2. How should the $180,000 be allocated, and what must be true before the forklift barrier is deferred?
Read the worked answer
Sequence the decision
- Re-score first. A forklift-pedestrian struck-by in a 60-trip-per-shift aisle has fatality potential; a severity of 3 is almost certainly a scoring error, not evidence of low risk.
- Move compliance-driven and feasible engineering controls to the front: auger cleanout guarding and lockout (machine guarding and energy-control requirements apply regardless of matrix score, and a bypassed or non-existent lockout is an imminent hazard).
- Fund the press-brake light curtain; recurring point-of-operation exposure with a known engineering fix is not a candidate for deferral.
- Phase the barrier, but install interim separation now: dedicated pedestrian route, physical railing, speed limits, mirrors, spotters, and a traffic plan.
ALARP discipline A deferral must be justified by gross disproportion between cost and risk reduction, documented, time-bound with an owner and date, and approved above the person who owns the production schedule. ALARP never licenses deferring a legally required control. Allocate by risk reduction per dollar and fatality potential, not by raw score.
3. Name two leading indicators that would show whether the changes are working, and explain why they are better than TRIR here.
Read the worked answer
Candidate indicators
- Percent of observed cleanouts with verified energy isolation and a stored-energy check (condition and behavior audit, not a records check).
- Near-miss and error reports per 100 maintenance hours, tracked for reporting trend rather than raw count.
- Percent of risk assessments completed with team scoring and re-scored within 30 days of a change.
- Corrective-action verification rate at 30/60/90 days.
Why these TRIR at 640 employees is a low-count, high-noise statistic; a single case swings it, and it reports harm after the fact. These indicators are frequent, tied directly to a specific control, and actionable within weeks. They also expose gaming: a lockout audit that only counts permits issued will look good while jumpers stay in the toolroom, which is why observation of actual task performance matters.