By Luke Martin – Partner and Safety COI Lead, ScottMadden; co-written by Alex Tylecote and Matt Reed
The Problem: Field Safety Is Largely Reactive
For decades, utilities have measured safety performance primarily through lagging indicators, most notably the Total Recordable Incident Rate (TRIR). While TRIR is required for compliance and remains a visible benchmark for executives and regulators, it is a backward-looking measure. It tells us what has already happened but fails to capture the conditions that predict future incidents.
Research from the Construction Safety Research Alliance (CSRA) has revealed the scope of this problem: because serious incidents are statistically rare, TRIR calculations swing widely with even small changes in recordables. A utility with two incidents in one year and three the next might see its TRIR spike by 50%, creating a misleading perception of worsening safety performance. However, with such small numbers of incidents for a single utility, TRIR is often statistically insignificant. Focusing on TRIR for compliance purposes may distract utilities from the fundamental drivers of risk.
This matters. Serious injuries and fatalities (SIFs) often stem from precursors that differ entirely from those behind minor injuries. Lack of planning, missing controls, or overlooked hazards are potentially relevant signals, but they remain largely invisible or ignored. As a result, even as utilities digitize more safety data, serious incidents have not declined proportionally.
Two barriers explain the gap between data collection and improved outcomes:
· Siloed information: Paper-based systems, personal notes, and informal communications prevent a full view of risk. Even where digitization has occurred, much of the most valuable information remains locked away in narrative fields or PDF attachments.
· Analytical limits: Traditional systems can tabulate incident counts or calculate rates, but they cannot process unstructured data or connect disparate signals into a meaningful risk picture.
To truly advance safety, utilities must look beyond compliance metrics and adopt approaches that surface leading indicators before harm occurs.
The Shift: How AI Makes Safety Predictive
Artificial intelligence offers a path forward. Unlike traditional analytics, which require structured datasets, large language models (LLMs) and other AI methods can process the messy, real-world inputs where SIF precursors often hide: handwritten field observations, free-text incident narratives, transcribed toolbox talks, and even voice recordings.
By analyzing this data holistically, AI can:
· Identify high-risk work conditions, such as jobs involving heights, electricity, or confined spaces, especially when proper controls are missing.
· Recognize risk patterns, like frequent overtime combined with hazardous tasks.
· Detect subtle but predictive signals, such as persistent ergonomic discomfort, that may escalate into recordable injuries.
For example, a recent benchmarking study showed that when an employee reported discomfort lasting more than 72 hours, the likelihood of a recordable injury increased to 75%. AI enables safety teams to act on these kinds of signals early, preventing harm rather than responding after the fact.
The shift is profound: LLMs enable utilities to broaden their scope, capturing a more comprehensive picture of field conditions and transforming disparate data into actionable insights. Instead of asking "What went wrong?" utilities can begin asking “Where are we most at risk, and how can we prevent it?”
On the Frontlines: Emerging Tools in Action
Utilities are beginning to pilot LLM tools across safety contexts. While most deployments are still early stage, these examples show the breadth of possibilities:
· Immersive Training & AI Risk Assessment: A leading battery manufacturer piloted immersive, video-based training “dojos” and AI-driven risk scoring tools. These systems highlighted hazards before work began and reinforced safer practices. In early results, participants demonstrated stronger hazard recall and greater confidence in applying controls in the field.
· AI-Enhanced Field Training: Our NXT GEN® Training platform has shown how virtual simulations of high-risk scenarios, combined with AI monitoring, can strengthen safety behaviors. For example, electrical workers practiced lockout/tagout procedures in a virtual environment, while AI can track common errors and flag areas for coaching.
· Visual Analytics at Scale: Transmission and distribution utilities are using AI to process drone footage, field photos, and vehicle camera feeds to identify equipment damage, vegetation encroachment, PPE adherence, or unsafe practices. One utility reported that AI-assisted drone inspections reduced manual review time by 70%, while also identifying potential hazards that were previously missed in traditional walkthroughs.
· Predictive Job Planning: Some utilities are piloting AI systems that score each upcoming job's risk level. Supervisors then decide whether to proceed, pause, or add additional controls. In one
pilot, supervisors received automated prompts when job risk scores exceeded a threshold, helping them reprioritize work schedules to lower exposure.
· Voice-Enabled Safety Assistants: AI-powered assistants allow field crews to query procedures hands-free, improving access to guidance during high-risk work. Imagine a lineworker about to enter a confined space asking, "What are the top three controls I need in place?" and receiving immediate, accurate guidance.
These tools are not intended to replace human judgment. Instead, they extend and augment it by providing frontline crews and managers with richer, real-time insights that enable smarter, safer decisions.
Predictive Safety in Practice: Dashboards and Metrics
Forward-looking utilities are pairing AI with new safety KPIs that emphasize leading indicators. These dashboards provide managers with a dynamic view of risk across crews, sites, and various time horizons.
Lagging Indicators (Traditional)
· TRIR (Total Recordable Incident Rate)
· Lost Time Incident Rate
· Serious Injuries & Fatalities (SIFs)
· OSHA-reportable incidents
· DART (Days Away, Restricted, or Transferred)
Leading Indicators (Predictive)
· Near misses reported
· Ergonomic flags raised
· Safety observations submitted
· Supervisor-led field coaching sessions
· Leadership safety walks
· Job briefs with AI risk alerts
By analyzing both types of data together, utilities can see patterns they would otherwise miss. For example, an uptick in near misses, combined with increased overtime hours, might predict an elevated SIF risk in a particular division. AI-powered dashboards make these connections visible in time to act.
The shift toward leading indicators is also reshaping culture. Instead of waiting for incidents to measure progress, utilities are rewarding proactive behaviors, such as reporting near misses, raising ergonomic concerns, and conducting peer coaching. These actions, amplified by AI insights, are redefining what constitutes "good safety performance.”
The Organizational Challenge: Beyond the Technology
AI's potential in safety is real, but adoption requires more than technology. Utilities must:
· Digitize data sources - Make handwritten notes, paper forms, and fragmented systems accessible for analysis.
· Foster leadership buy-in - Encourage decision-makers to trust and act on AI-generated insights.
· Manage change - Ensure crews see AI as a tool that supports, not replaces, their expertise.
· Balance imperfect data - Recognize that perfect data is not required to begin; rapid prototyping can validate AI concepts even with incomplete information.
Cultural readiness may be the most significant barrier to effective communication. Crews need to trust that AI is not about surveillance, but about protection. Leaders must be willing to act on AI insights, even when they challenge traditional assumptions. Building that trust requires transparency, training, and consistent communication.
Utilities with strong safety cultures and innovative mindsets are already moving forward. Those waiting for flawless digital infrastructure risk falling behind.
A Roadmap for Utilities
Based on industry pilots and lessons learned, utilities can approach AI adoption in three phases:
1. Discover and Prioritize - Identify high-value safety use cases, assess data readiness, and quantify potential impact. For example, start with jobs known to involve high-energy hazards and evaluate how AI might surface early warning signs.
2. Pilot and Learn - Launch small-scale pilots to validate AI's effectiveness in specific contexts, building stakeholder confidence through demonstrated results. Early pilots might focus on a single hazard category, such as ergonomic strain, or a specific operational area, like substation maintenance.
3. Scale and Sustain - Develop enterprise-wide roadmaps, embed AI into processes, and establish governance to ensure responsible, long-term use. Success requires not only technology but also training, change management, and leadership alignment.
This phased approach enables organizations to transition from exploration to impact without overcommitting resources too early. It also creates space to adapt via learning from pilots, refining models, and scaling what works.
The Broader Impact: Why This Matters
The benefits of predictive safety extend beyond injury prevention:
· Operational resilience - By reducing incidents, utilities minimize unplanned outages and disruptions.
· Workforce engagement - Crews see leadership investing in tools that keep them safer, strengthening trust and morale.
· Regulatory alignment - As regulators increasingly emphasize leading indicators, AI-powered safety programs can help utilities stay ahead of compliance trends.
· Reputation and recruitment – Companies with strong safety cultures are more attractive to prospective employees, especially as the industry faces workforce transitions.
What's Next
The shift to predictive safety is happening now. Early adopters are already seeing results, and their successes will soon become industry benchmarks. Utilities adopting AI-enabled safety analytics are setting the standards that others will quickly follow. Those that remain reliant on lagging indicators risk falling behind in both performance and reputation.
Importantly, AI amplifies human expertise. A safety manager who once spent hours reading reports can now receive AI alerts highlighting emerging risks. Supervisors can tailor safety briefings to real-time conditions. Crews can access guidance in the field when they need it most. The combination of human judgment and AI insight makes safety systems stronger, not weaker.
The next step is clear: utilities should experiment, pilot, and share lessons. Together, the industry can define how predictive analytics reshape field safety for the better.