Overview
This project sits at the intersection of AI, enterprise operations, and human judgment. I led the work that turned an early predictive model into a usable product direction for field supervisors making safety-critical decisions under time pressure.
Principal Design Lead responsible for vision, strategy, research, systems framing, workflow design, POC experience design, and executive readout. I led the POC and helped shape the pilot direction.
Stakeholder interviews
user research
rich picture
journey map
service blueprint
app and user flows
design requirements
data feature prioritization
wireframes
visual design, and executive presentation
POC prototype
Critical crew assignment decisions were being made through fragmented tools, paper, spreadsheets, and local workarounds. The organization had an early safety prediction concept, but not a clear product strategy, workflow model, or trusted user experience.
Reduce injury and fatality risk
Improve leadership visibility into safety decisions
Enable more proactive risk management
Test the viability of embedding predictive intelligence in the daily workflow
A digital Crew Board that embedded a risk calculator and risk drivers into the existing Morning Shuffle planning workflow. The system surfaced elevated-risk work orders, highlighted key drivers, and supported action without automating the final decision.
The Safety Predictive Model analyzed 200+ variables per work order, and the Digital Crew Board reached 80% accuracy in flagging elevated-risk work orders
The platform was rolled out across all 35 districts
During the broader deployment period, the organization reported a 62% reduction in contractor serious injury and fatality rate between 2018 and 2021, even as contractor hours grew 35%
In 2022, the organization also reported a 62.7% improvement in public serious injuries and fatalities versus its five-year average
Recognized with a National Safety Council Innovation Award in project materials
What, Why and How
I led the shift from “we have a predictive model” to “we have a product people can trust in a live safety workflow.”
This mattered because supervisors had 30 to 45 minutes each morning to assign crews, review work orders, and send people into the field. The design challenge was not how to show more data. It was how to give the right intelligence at the right moment, without removing human accountability.
Mission
The mission was to redesign a safety-critical workflow, not just improve a screen.
A large electric utility needed a way to identify potentially risky work before crews were dispatched. My role was to turn that intent into a product direction that could support field supervisors, align leaders, and make the AI usable in the real world through a proof of concept and pilot direction.
Problem
This was a high-stakes, complex workflow with low trust in digital tools.
Across districts, supervisors relied on paper logs, spreadsheets, fragmented communication, and an outdated scheduling tool. Safety data was disconnected from the moment decisions were made. Leadership wanted more proactive risk management, but the path from predictive model to daily operational use was still unclear.
Early Insights
The real problem was the workflow
The model alone was not the product. The real design problem was the crew assignment ritual around it.
Trust was a product requirement
Supervisors already had a mental model that worked for them. If the new system felt unfamiliar or intrusive, adoption would fail.
AI needed to support judgment, not replace it
In this context, automation would have weakened trust. Decision support was the right pattern.
Alignment required visible systems thinking
Rich pictures, journey maps, and service blueprints were not presentation artifacts. They were how I helped stakeholders see the true shape of the problem.
- Digitize a heavily manual planning workflow without disrupting critical daily operations
- Embed predictive risk insight into the exact point of crew assignment
- Give supervisors clear, timely signals without overwhelming them
- Build enough confidence across business, data, engineering, and leadership to move from idea to pilot
Designing AI, not just advocating for it
My approach here was AI-native, even though the system was built around predictive intelligence rather than generative AI.
I did not ask, “Where can we add AI?” I asked, “If the model can reduce the cognitive load of this decision, what is the human’s role now?” That shift changed the product.
the AI flags elevated risk the system explains the signal through top drivers the supervisor keeps decision authority the workflow captures action in context
the system explains the signal through top drivers
the supervisor keeps decision authority
the supervisor keeps decision authority
I started by mapping the current state across people, tools, handoffs, and failure points. That work exposed how much of the process was still manual and where safety risk was being managed too late.
There was early resistance to user research. I kept pushing because the team’s understanding of the field reality was incomplete. Interviews with district managers and crew leaders changed the discussion from abstract innovation to operational fit.
The challenge was not adding a score to a screen. It was designing a decision-support experience that fit the planning ritual supervisors already used. I designed flows that surfaced elevated-risk work orders, showed top drivers, and placed that insight inside the crew assignment workflow.
The interface intentionally mirrored the physical and mental structure of the existing Morning Shuffle. Familiar labels, recognizable groupings, and clear visual hierarchy helped the product feel usable on day one.
I led the POC from problem framing through experience design and executive presentation, then helped shape the direction of the pilot as the concept moved closer to real-world use. That boundary matters because it shows strategic leadership without overstating delivery ownership.
That is the pattern I still use in enterprise AI: AI at the decision moment, human judgment at the point of accountability.
The Result: POC Prototype
This case study earns its place in a 2026 and 2027 portfolio because it shows the core capability many AI-first companies still struggle to find: the ability to turn a promising model into a trusted operating system for real work. It demonstrates systems thinking, executive clarity, Human-in-the-Loop design, and hands-on product craft in a context where the stakes were real.
I designed around the existing planning ritual so the product could be adopted. A more radical interface would have been harder to trust.
Supervisors needed enough reasoning to trust the signal, but not a dense analytics view that slowed them down.
The pilot did not replicate the POC exactly because it had to work within platform constraints. The goal was to preserve the decision-support logic, not every visual detail.
The strongest outcome story here is organizational, not single-feature attribution.
The Safety Predictive Model and Digital Crew Board brought predictive risk insight into daily crew assignment across 35 districts, with 80% accuracy in flagging elevated-risk work orders. During the same deployment period, the organization reported a 62% reduction in contractor serious injury and fatality rate between 2018 and 2021, even as contractor hours grew 35%, followed by a 62.7% improvement in public serious injuries and fatalities in 2022 versus its five-year average.
Daily Morning Shuffle, 30 to 45 minute planning window, crew assignments before field deployment
Risk calculator integrated into Crew Board, top risk drivers, Human-in-the-Loop logic, recommendation without auto-assignment
Rich picture, journey maps, service blueprint, app flows, future-state ecosystem
Executive readout, workshops, alignment across business, Salesforce, solution, and design partners
Materials cite 35 districts
National Safety Council Innovation Award cited in materials
This project reinforced that enterprise AI succeeds when the workflow changes, not just the interface.
It also confirmed that trust is built through timing, clarity, and role definition
In high-stakes environments, the best AI products do not take control. They reduce cognitive load and make human judgment better.
It also reminded me that research is often the fastest path through ambiguity, especially when stakeholders think they already understand the problem.
This also sharpened how I talk about impact in enterprise AI: the most credible claim is not that one interface changed everything, but that a well-designed decision system helped operational teams act earlier, with better visibility, across a much larger safety transformation.
