AGV Commissioning at NorthBridge Cold Chain
NorthBridge Cold Chain operates a 12,000 m² refrigerated warehouse and is nine weeks from commissioning 18 automated guided vehicles (AGVs) for pallet movement. The risk register contains 28 entries, dominated by 'AGV equipment failure' and 'cyber attack'; most were copied from the vendor's generic list. Three weeks ago an AGV clipped a rack upright at 02:10, destroying six pallets and closing an aisle for four hours. The night-shift supervisor had noticed erratic routing a week earlier but did not file a report because 'nothing happened'. The project board proposes to score all risks using a 5x5 matrix populated only with the site's last three years of incident data (four recordable incidents). The contingency is 8% of a £1.4m budget. The operations director wants the register finalised in five days.
1. Identify the risk-identification gaps and cognitive biases evident in the register and near-miss handling, and explain how you would close them before any scoring takes place.
Read the worked answer
The register is not site-specific. 'AGV equipment failure' and 'cyber attack' are broad labels, not scenarios, and the vendor list will miss layout-dependent risks: aisle width, rack protection, floor markings, charging-area fire load, Wi-Fi dead zones, pedestrian/AGV interfaces, night-shift manning, and manual override. The near-miss shows an outcome bias: the supervisor judged reportability by the result ('nothing happened') rather than by the potential for harm. That is also normalisation of deviance, because erratic routing was tolerated for a week. Availability bias appears in the board's preference for incident data: four recordable incidents are easy to recall, while the AGVs have almost no operating history. Confirmation bias may explain continued reliance on the vendor list.
Before scoring, I would close the gaps with structured identification: run a HAZOP or what-if workshop across AGV operating modes (normal transfer, charging, manual recovery, network loss, congested aisle); walk the floor at night; interview night shift and maintenance; review vendor FMEA and site layout; and add human-factors and commissioning/change risks. I would also issue a simple near-miss definition and a no-blame reporting route, because a reporting rate of zero after a visible routing deviation is a warning sign, not assurance. Only after the register contains causes, events, consequences, existing controls, and owners should the team score likelihood and consequence on defined scales.
2. Evaluate the board's proposal to use only historical incident data with a 5x5 matrix. Recommend an analysis approach for the AGV risks that is proportionate to the nine-week window and 8% contingency.
Read the worked answer
The proposal is not sufficient for the decision. Historical data has very low statistical power: four recordable incidents in three years, none involving AGVs, cannot estimate collision, rack-strike, or telemetry-loss risk. A 5x5 matrix can rank risks for discussion, but it will not reveal causes, control gaps, or cumulative exposure, and it is vulnerable to inconsistent scales and range compression. Using it with sparse data creates false precision.
With nine weeks and 8% contingency (£112k), I would use a tiered approach. First, qualitative HAZOP/what-if for all AGV modes to identify deviation scenarios. Second, FMEA for the higher-consequence scenarios: AGV strikes rack or person, load loss, uncommanded movement, charging fire, and loss of telemetry. Third, a bow-tie or fault tree for 'AGV strikes rack/person' to connect preventive controls, detection, and mitigation. Fourth, collect leading indicators during trials: route deviations per 1,000 trips, emergency stops, near-miss reports, Wi-Fi dropouts, and manual overrides. Fifth, use vendor mean-time-between-failure data only as a cross-check, with a scenario range or simple Monte Carlo for downtime and financial loss. The contingency should fund supervised night trials and independent review, not just scoring workshops. This gives a defensible register within the deadline.
3. The near-miss was not escalated. Evaluate the risk criteria and escalation arrangements you would set, including specific triggers and reporting expectations.
Read the worked answer
The current criteria are effectively 'report if something bad happened', which failed. I would define a near-miss as any unplanned AGV event with potential for harm or loss, regardless of outcome. Specific triggers should include: any AGV deviation within 0.5 m of a rack upright or person; any uncommanded movement; any telemetry or Wi-Fi loss over 60 seconds; any manual override in a live aisle; any emergency stop; and any repeated routing exception. These are observable and testable.
Reporting expectations: stop and isolate the vehicle where safe, notify the shift manager within one hour, enter the event in the reporting system within 24 hours, and have the risk owner review within 48 hours. Escalation to the project board and safety committee should occur within 24 hours if there is actual contact with a rack or person, if the same cause repeats twice in 30 days, or if a control is found not to be working. The form should capture what happened, what prevented harm, the state of controls, and immediate action, using a no-blame process. The board should track reporting rate as a positive indicator, not just incident count. This replaces vague 'significant event' language with triggers that night shift can apply at 02:10.