Your EHS data is trying to tell you something. Are you listening?

I started my career in EHS in 1995. I’ve watched the industry evolve from paper-based corrective action logs to sophisticated digital systems. And now we’re at what I believe is the most significant inflection point in that entire span: AI that can read the unstructured text inside your data and surface patterns you’d never find manually. 

But it only works if your data is honest. That was the thread running through my session at Safety on the Edge—and the conversation that followed. 

The problem hiding in plain sight 

For years, on-time corrective action closure has been one of EHS’s most-watched leading indicators. That pressure created a behavior most of us recognize but rarely say out loud: pencil whipping. Actions entered to meet the deadline, not solve the problem. Root cause entries that default to “employee needs retraining” because it’s the path of least resistance. 

Our data shows about 70% of corrective and preventive actions are closed on time across our customer base. That sounds solid. But it tells you nothing about whether the actions were actually effective. When you layer AI on top of historical data, those gaps become visible quickly. 

What AI found in 4,400 corrective actions 

I walked through a real example during the session. A customer in Pennsylvania had 4,400 corrective actions on record, pulling from their incident module, audit module and compliance calendar. We leveraged our AI Agents to assess the details of the corrective actions to find patterns beyond traditional drop-down categories. 

Around 300 actions were tied to emergency response equipment blockage. The biggest single pattern? Eyewash station obstructions. 

This is a company running at a 0.28 incident rate: essentially world class. But buried in their data was a signal they hadn’t caught: basic safety fundamentals being overlooked on the floor. Their response was immediate: a back-to-basics campaign for the year. Not because AI told them what to do, but because it surfaced what they already had and couldn’t see. 

AI accelerates thinking - it doesn’t replace it 

There was a sharp question from the audience: how do you stop AI from just becoming a faster way to pencil whip? It’s the right concern. Our answer was to build in friction. When AI suggests a 5 Why analysis or recommends a corrective action, the user still has to review it and enter it manually. The system doesn’t populate fields on their behalf. 

We also built an effectiveness validation step: once a corrective action is closed, AI assesses the quality of the closeout against the original source. That’s a fundamentally different signal than on-time closure—and much harder to game. 

Where to start 

AI review boards are real. We’re seeing organizations take six to nine months to get internal approval to run EHS data through AI tools. My advice: start small, keep it internal and focus on data quality first. AI is only as useful as what you give it. If years of corrective action entries were made to hit deadlines rather than document what actually happened, the patterns AI surfaces will reflect that. 

The shift from reactive to proactive safety management isn’t a technology problem. It’s a data problem that AI is finally equipped to help you solve—provided the data is worth solving with. 

---

Not sure where your data stands? 

Before investing in AI, it’s worth knowing whether your incident reporting gives you an accurate picture. Ideagen’s free checklist, 7 signs your incident reporting is giving you a false sense of security, walks you through the most common signals that your data quality may be masking real risk—including low near-miss rates, generic root cause entries and corrective action backlogs that nobody trusts.