It is Sunday night. The quarterly business review is on Tuesday afternoon. The board materials are due by end of day Monday. The slide you are working on says something like, “EHS performance by business unit, Q3,” and it is the slide that always generates the hardest questions.
You know the questions you will be asked, because you ask them yourself when you are looking at the data. Which business units are trending in the wrong direction? Where are the leading indicators softening before the lagging indicators show it? Is the drop in work observation participation in the Northern region the reason their incident rate is creeping up, or is something else going on?
These are the right questions. They are also, until this week, questions you could not answer on Sunday night. They required an analyst, an Excel export, two or three iterations to get the joins right, and probably a phone call to the platform vendor. The honest answer at the board table has often been some version of “let me follow up on that one.”
That changes with EHS Insight’s new MCP connector. The questions you used to defer to follow-up email become questions you answer in the room.
Every EHS platform on the market, including ours, has dashboards. Dashboards are excellent at answering one kind of question. What. What is our recordable rate. What was last quarter’s incident count. What is the audit completion percentage.
They are not built to answer the questions a director actually fields. Why. Where. Which. Is there a correlation. Show me the entities trending in the wrong direction, and show me what they have in common.
The reason is structural. A dashboard is a pre-built view of a pre-defined slice of data. Every slice that anticipates a question gets built; every slice that does not, does not exist. The question that has not been anticipated requires a custom report. The custom report requires an analyst. The analyst requires time. And so the strategic question, the one that would actually move your program forward, gets deferred.
That deferral has been the rate-limiting factor on how strategic EHS leadership can be. It is not a question of intent. It is a question of access to the data.
With MCP, you open Claude, ChatGPT, or Microsoft Copilot. You type the question your board is going to ask you on Tuesday.
|
“Which business entities have seen worsening incident rates? Which business entities have seen worsening Work Observation participation? Is there a correlation?” |
|
|
The AI plans the analysis. It pulls incident rate trends from the relevant time window. It pulls observation participation trends from the same window. It identifies the business units that show declines on both. It surfaces the correlation. It presents the answer in a format you can use, with the entities listed and a short narrative summary explaining what the data shows.
Ask one more question. “Turn that into three slides for Tuesday’s board meeting. Use the format from last quarter’s deck.” The AI drafts the slides. You review, edit, and finalize. The work that used to take an analyst plus a back-and-forth cycle becomes a half-hour conversation that ends with the deck closer to done.
“A safety manager can pull live EHS Insight data out into that same conversation where they are already drafting an executive summary or replying to an email or updating specific slides in a PowerPoint that already exists.”
-Eric Stevens, Chief Technology Officer, EHS Insight
The strategic query above is a useful starting point because it is the kind of question that would have been on a dashboard wishlist that never got built. The more interesting territory is the second-order question that follows.
Once the AI surfaces the correlation, you can keep asking. “Let’s look at the three units where the correlation is strongest. What do they have in common? Same leadership? Same operational changes in the last six months? Same customer pressure? Pull whatever signal you can find.”
This is the conversation an EHS director runs in their head while looking at the data, except it never makes it into the analysis because there is no time and no analyst on standby. With MCP, that conversation becomes the analysis. The director becomes the analyst, with the AI as the technician.
It is worth saying what this is not. It is not a magic answer machine. The AI is not generating insights out of nothing. It is reading the data you already have, in EHS Insight, with the relationships and structure your team already built, and applying analytical reasoning on top. The intelligence in the answer is your data, surfaced fast enough to be useful.
Every other AI play in our category is a chatbot. Their AI lives in their product. Their AI answers questions in their format. Their AI gets better when their engineering team makes it better, on their release cycle. Our customers do not get to choose which AI they prefer. They get whichever one their EHS vendor built.
MCP is structured the opposite way. We built on an open standard. The AI you use for everything else, the one your security team has reviewed, the one your team has standardized on, that is the same AI that talks to EHS Insight. When that AI gets better, your EHS analysis gets better. When Claude releases a new model, when ChatGPT improves at multi-step reasoning, when Microsoft Copilot expands its enterprise capabilities, you do not have to wait for us to ship anything. You get the upgrade automatically.
For an EHS director thinking about a three-year horizon, this is the more important point. The questions you can ask of your program in 2026 are good. The questions you will be able to ask in 2027 and 2028, on the same connector, will be better, because the AI itself will be better. The investment compounds.
In a year, the EHS director who is fluent with the MCP connector will run a different kind of program. Strategic questions will be asked in real time, not in follow-up email. Correlations will be checked before they become problems. Board meetings will be more pointed because the data is ready, not waiting on a deck.
The director’s role will shift slightly, too. Less time pulling threads through the data, more time deciding what to pull threads on. The analyst work becomes table stakes. The strategic work becomes the differentiator.
That is the right direction for the EHS function. It is the direction safety leadership has been promising for ten years, in conference keynotes and trade press editorials. MCP is one of the things that finally makes it operational.
How is this different from the BI tool we already have plugged into our EHS platform?
BI tools answer pre-built questions. They are excellent at “what is the recordable rate” and “show me the incident count by site.” They struggle with “which units are trending wrong on both incidents and observations” because that requires correlation across modules and the construction of an analysis on the fly. MCP gives your AI tool the ability to do that construction conversationally, without a custom report.
Do I lose access to the dashboards I already use?
No. Dashboards, scheduled reports, and the EHS Insight UI continue exactly as they are. MCP is an additional channel, not a replacement.
What happens to the analyst on my team?
Their work changes. The reformatting and ad-hoc data pull work that currently consumes a meaningful portion of their week moves to the AI. The strategic work they have always wanted to do but rarely had time for becomes their main job. Most EHS leaders we have spoken with describe this as a relief for their analysts, not a threat.
What if our AI tool is different from Claude, ChatGPT, or Copilot?
MCP is an open standard. The three AI assistants named at launch are the ones currently supporting the protocol. As the standard sees broader adoption, support extends automatically. You are not locked into a specific AI vendor through this integration.
How do I trust the answers the AI gives me?
The AI is not generating answers from its training data. Every answer comes from your live EHS Insight records, filtered by your permissions, returned through the same API that powers the EHS Insight UI. The AI’s job is to translate your question into the right query and present the result. The data is real. The analysis is grounded.