The Near-Miss That Nobody Logged
A mid-sized logistics firm runs 340 vehicles. Over 14 months, the safety team recorded 9 minor incidents (wing-mirror strikes, low-speed reversing knocks). Drivers privately describe at least 30 more near-misses that were never reported, mostly during the 04:00-06:00 shift change. The firm's risk register lists 'vehicle collision' as a single line item, rated Likelihood 2 / Impact 4, unchanged for three years. The newly hired risk manager notices the register has never been reviewed against the incident log, and that drivers fear disciplinary action if they report a near-miss. The operations director says: 'Nine incidents in 14 months is a good safety record - the register rating is fine.'
1. The operations director concludes the register rating is validated by the low recorded incident count. Evaluate this reasoning and identify what the data actually indicates.
Read the worked answer
The director is treating an absence of recorded events as evidence of low likelihood, but the reporting system is itself compromised. Three signals undercut the rating:
- Reporting suppression. Drivers fear discipline, so near-misses go unreported. The count of 9 measures reporting behaviour as much as it measures collision frequency.
- Unobserved population. At least 30 disclosed near-misses against 9 recorded incidents means the observed rate may understate the true rate by a factor of roughly four.
- Concentration. Near-misses cluster at the 04:00-06:00 shift change, which points to a specific causal driver (fatigue, handover) rather than a diffuse background rate.
A Likelihood of 2 held constant for three years is not a validated rating; it is a stale one. The correct conclusion is that the rating is unverified and probably understated, and that the reporting culture is the first thing to fix.
2. Recommend how the firm should restructure its treatment of this risk, distinguishing between what changes in the register and what changes in the organisation.
Read the worked answer
Register changes: Break the single 'vehicle collision' line into distinguishable risk events - e.g. reversing incidents in depots, fatigue-related incidents at shift change, and highway collisions. Each gets its own likelihood, impact, and owner. A single blended line cannot be managed because the controls differ.
Organisational changes: Introduce a blame-free near-miss reporting channel and measure reporting volume as a positive indicator rather than a failure. If reports rise, that is evidence the control is working, not that safety is worsening.
Sequencing: Fix reporting first, then re-rate likelihood once a few months of honest data exist. Re-rating now would just swap one guess for another.
Governance: Give the register a review cadence tied to incident data, not to the calendar alone, so it cannot sit unchanged for three years again.