Case study 04 · Granite Construction · Jun–Jul 2026
Light-Vehicle Incidents & Field Work
Moving a supervisor conversation from driver blame to exposure management, and the jobsite hours that grounded every data project I touched.
- My role
- Analyst and presenter for the light-vehicle review; jobsite inspections and safety observations; the link between a corporate AI team and the field.
- Users
- Regional supervisors at a SHEQ (safety, health, environment, quality) meeting; crews on active sites; the corporate team building safety AI.
- Methods
- Incident review · pattern analysis · jobsite inspection and observation · SHEQ assessment participation · stakeholder alignment · presenting to supervisors
- Scope
- Every documented light-duty vehicle incident in the region in 2026; five active sites visited; one project SHEQ assessment.
- Timeline
- June to July 2026, presented at the mid-summer supervisor meeting.
- Outcome
- Four low-cost, supervisor-led corrective actions adopted as the recommendation; a standing bridge between the AI team and field users.
Why do our trucks keep getting hit?
Light-duty vehicles, the pickups and crew trucks that get people to the work, were showing up in the incident log more than anyone liked. The default explanation was driver behavior. I read every 2026 light-vehicle incident in the region to see whether the reports agreed.
Struck while parked
Unattended vehicles left near active equipment or public traffic. Nobody was driving.
Tight-workspace congestion
Light vehicles and heavy equipment sharing the same narrow routes.
Low-speed contact
Backing in confined spaces; contact with fixed objects at walking pace.
Third-party factors
Subcontractor equipment and public vehicles inside the contract limits.
02 · What I recommended
Four controls a supervisor can put in place this week
The data pointed to fixes within supervisory influence and low-cost to implement, with no budget or process overhaul required. That framing mattered: a supervisor meeting is the wrong room for a policy proposal and the right room for something they can do on Monday.
01
Designated parking zones
Parking assigned away from active equipment and traffic to remove unattended-vehicle exposure.
02
Jobsite layout
Separate light-vehicle routes from heavy equipment with cones and barriers.
03
Targeted awareness
Look-before-you-move habits and confined-space hazard recognition, applied at the task level.
04
Subcontractor coordination
Housekeeping and coordination controls for third-party equipment contact.
Getting on site was not optional
Every classification rule I wrote that summer was checked against something I had seen. I conducted inspections and logged positive safety observations across five active sites, from a highway resurfacing job to a mechanic's shop, and took part in a full project SHEQ assessment alongside the regional safety leads.
Field time changed the work directly. Watching where crew trucks actually park is where the light-vehicle findings came from. Standing next to a lane closure is why the intrusion definition distinguishes a breached device from a public car that simply drove past. And the safety huddle times, PPE rules and meeting points I kept asking for on every visit became the data model for the Job Site Hub.
- 5active sites inspected, with positive observations logged for each
- 1project SHEQ assessment, conducted with the regional safety lead and the project's safety manager
- 3data projects whose logic was shaped by something observed on site
- 0classification rules written without having seen the hazard in person
The people building safety AI hadn't met the people who would use it
While requesting expanded access to build the intrusion classifier, I discovered that a corporate AI team was actively building safety tools with no contact with the field users who would rely on them. Nobody had decided that; the teams simply sat in different states and different org charts.
I became the link: sharing what supervisors and crews actually needed, feeding real regional data into their scenario work, and meeting with the safety and AI leads to keep the connection going after my internship. It is the most UX-research-shaped thing I did all summer, and it never had a project name.
Know the room, and make the fix fit the person who can make it
A finding is only useful if the person hearing it can act on it. Framing incidents as exposure rather than blame gave supervisors something to do instead of someone to discipline.
Observation is a research method, not a field trip. The hours on site were the difference between rules that sounded right and rules that held up.
Sometimes the most valuable deliverable is an introduction.