Case Study · Safety Design · AI Integration

Fatigue
Management

A modular safety system designed to move field operations from static declarations to continuous risk awareness — combining self-assessment, real-time sensing, AI-assisted monitoring, and forward-looking scheduling integration.

RoleBusiness Analyst & System Designer
DomainSafety · Field Operations
Components4 (Production → Exploratory)
Solution Architecture — Four Layers
Detection Layer
In-Cab Retinal Monitoring Blink Rate & Eye Closure Wearable Physiological Data Sleep Quality Tracking
Assessment Layer
Self-Assessment Tool Weighted Scoring Logic AI-Assisted Hour Recording Continuous Rule Evaluation
Integration Layer
Project System (Job Details) Scheduling System (Upcoming Shifts) Cost Allocation Codes
Response Layer
Real-Time Alerts Escalation Pathways Forward-Looking Breach Detection Manager Notification
Context & Challenge

Fatigue as a
systemic risk

Fatigue was a recognised safety risk across field operations — particularly for mobile and vehicle-based roles where impairment has immediate physical consequences. The problem was not awareness. It was architecture.

The existing approach was fragmented, reactive, and largely compliance-driven. Staff completed self-assessments at shift start and the process ended there — no continuous monitoring, no early warning, no connection to scheduling data that could flag future risk before it materialised.

Rather than a single tool, the solution was approached as a system of components — some delivered into production, others prototyped or investigated to de-risk future investment.

The design challenge was to build something that could be incrementally delivered while preserving a coherent long-term architecture — so that each component added genuine value on its own and compounded with the others.

My Role

Business analysis and system design
Reverse-engineering the existing assessment tool
Proof-of-concept development
Integration thinking
Technology investigation
Policy translation
Solution Components

A modular system
delivered incrementally

Four components spanning the full maturity spectrum — from production-deployed improvements through to exploratory technology investigations. Each designed to stand alone and compound with the others as the solution matured.

Production
📋
Fatigue Self-Assessment Tool
Existing tool — rebuilt from first principles

A fatigue self-assessment tool already existed but was limited in its effectiveness. Rather than replacing it, I reverse-engineered the automated scoring and weighted assessment logic to understand precisely how it worked — and where it didn't.

Key Findings
The scoring logic was incomplete and inconsistently applied across different question categories — some risk indicators were underweighted
Assessment outcomes lacked clear, actionable responses — a "pass" didn't trigger any meaningful intervention even when borderline risk was present
Users could technically satisfy the assessment without engaging with it meaningfully — compliance without safety
Analysis created the foundation for improving both the assessment model and its downstream decision-making pathway
Pilot
👁️
In-Cab Retinal Fatigue Detection
Passive, continuous sensing while driving

I contributed to a pilot exploring in-cab retinal and eye-tracking technology designed to detect fatigue indicators in real time — including blink rate, eye closure duration, and gaze patterns — while a vehicle was being operated.

The technology was positioned as an assistive safety layer, not a disciplinary tool. The distinction mattered: field crews needed to trust it before it could be effective. Alerts and prompts were designed to prompt a rest stop, not generate an incident report.

Why This Matters
Removes reliance on self-reported data — people under fatigue are poor judges of their own impairment
Continuous monitoring fills the gap between shift-start assessment and end of a long day in the field
Passive sensing reduces user burden — no additional action required from the operator
Proof of Concept
🤖
AI-Assisted Continuous Work Hours & Fatigue Assessment
ChatGPT-based — policy rules embedded in daily hour recording

I designed and built a ChatGPT-based fatigue assessment proof of concept that embedded fatigue rules directly into the daily work-hour recording process — turning a compliance task into a live safety intervention moment.

Rather than a separate assessment at shift start, fatigue rules were evaluated continuously as hours were entered. Approaching a threshold triggered immediate feedback. Breaching one triggered escalation logic based on the specific policy rule violated.

→ Project System Integration

Retrieved job details using job numbers entered during time recording — ensuring hours were correctly allocated to cost codes and enabling task-specific fatigue rules (driving vs non-driving activities carry different thresholds).

→ Scheduling System Integration

Retrieved upcoming shift assignments from the scheduling system — enabling forward-looking detection of potential fatigue breaches before they occurred, shifting management from reactive to preventative.

Exploratory
Wearable Fatigue Indicators
Physiological data as an objective input layer

I investigated the potential use of wearable devices — rings and watches capable of capturing physiological indicators — as a trusted input layer for automated fatigue assessment. The focus was on data that correlated with actual impairment: sleep duration and quality, heart rate variability, and recovery and strain metrics.

The design question was not "can we collect this data?" but "how does trusted physiological data feed into automated assessments in a way that reduces user burden and improves decision quality?"

Design Considerations
Data ownership and privacy — employees must trust how physiological data is stored and used
Integration pathway into the AI assessment layer as a pre-populated input rather than a manual entry
Threshold calibration — physiological norms vary between individuals, requiring personalised baselines
The Core Shift

From static declarations
to continuous risk awareness

⚠ Before — Compliance-Driven
Single point-in-time assessment at shift start — no monitoring after
Scoring logic incomplete and inconsistently applied
"Pass" outcomes without meaningful intervention thresholds
No connection to scheduling data — fatigue risk in upcoming shifts invisible
Hour recording separate from fatigue assessment — no integration
Reactive — problems identified after the fact, if at all
Entirely dependent on self-reported data
✓ After — Safety-Driven
Continuous assessment as hours are recorded throughout the day
Rebuilt scoring logic with consistent weighting and clear outcome pathways
Threshold breaches trigger automatic escalation and manager notification
Scheduling integration enables forward-looking breach detection — prevents rather than responds
Hour recording and fatigue rules unified in a single workflow
Passive sensing (retinal, wearable) provides objective data layer
Physiological indicators reduce reliance on self-assessment alone
Outcomes & Insight
🔍
Existing logic rebuilt from first principles
Reverse-engineering the assessment tool revealed gaps that had existed since the tool was built — scoring inconsistencies and missing escalation logic that meant real risk could pass undetected. The rebuild gave the tool the foundation it needed to be effective.
🤖
AI turned compliance into intervention
The ChatGPT proof of concept demonstrated that embedding fatigue rules into daily hour recording — a task people already do — could turn a routine activity into a live safety check with real-time feedback, without adding meaningful burden.
📅
Scheduling integration shifted the model
Connecting the assessment layer to upcoming shift data was the single most consequential design decision — it changed fatigue management from a question asked at the start of a shift to a continuous forward-looking risk model.
🏗️
Architecture designed for incremental delivery
The modular design meant components could be delivered in sequence — each adding value immediately and enabling the next. No single investment was required to realise benefit, reducing risk and making the business case for each component easier to approve.
🛡️
Safety culture, not surveillance
Every design decision kept the distinction between safety tool and disciplinary tool explicit. Trust is the prerequisite for effectiveness — field crews who don't trust a system will find ways around it.
🗺️
A roadmap for future investment
The exploratory work on wearables and the retinal pilot created a de-risked technology roadmap — evidence-based assessments of what each layer would deliver, what it would cost, and what needed to be true for each to succeed.