The EHS AI Readiness Checklist

1. When discussing requirements for a new EHS system, how is "AI" typically described?

(1 Point) As a single line-item in a list of features, like "Add AI for intelligent suggestions," with the expectation that the vendor will figure it out. (2 Points) As a high-level goal, such as "We want to use AI to predict risks," but without a detailed scope for the data or R&D effort required. (3 Points) As a distinct strategic initiative, with dedicated discussions about specific use cases, data requirements, necessary expertise, and long-term goals.

2. What is the primary business expectation for an AI solution within your EHS program?

(1 Point) We expect it to work "out of the box" to automatically find root causes or predict incidents with minimal human oversight, much like a magic wand. (2 Points) We expect it to improve efficiency by automating existing tasks, such as classifying incident reports or identifying trends from dashboard data. (3 Points) We expect it to be a long-term capability that evolves. We plan to start with a specific, well-defined problem and understand it will require ongoing investment and refinement to deliver strategic value.

3. How is the budget for AI in EHS being allocated?

(1 Point) The cost is assumed to be a small, inclusive part of the overall EHS software license fee. (2 Points) We've allocated a budget for an "AI module" from a software vendor, but have not yet budgeted for the internal resources needed for data preparation, validation, and training. (3 Points) We recognize it as an R&D-level investment and have a dedicated budget that accounts for both the technology and the involvement of our internal subject matter experts.

4. How would you describe your organization's historical EHS incident data?

(1 Point) It is decentralized in various spreadsheets, Word documents, or legacy systems with inconsistent formats and terminology. It's essentially "crude oil." (2 Points) We use a central system, but data entry practices are inconsistent across different sites or business units. The same term (e.g., "Lost Time Incident") can have different meanings. (3 Points) We have a formal data governance policy. Data is captured in a structured, centralized system with clear, enforced definitions and is subject to regular quality audits.

5. How are subjective EHS assessments (e.g., risk severity, likelihood) handled across your organization?

(1 Point) They are based purely on the discretion of individual managers, resulting in significant variation from one report to the next. (2 Points) We provide general guidelines and training on how to use our risk matrix, but we lack a systematic process to ensure it's applied consistently. (3 Points) We use a standardized, calibrated methodology (e.g., a detailed risk matrix with explicit criteria for each level) that is enforced and audited to ensure consistency.

6. If you were asked to provide a "clean" and "labeled" dataset for an AI proof-of-concept today, what would that process look like?

(1 Point) It would require a massive, manual effort to find, consolidate, and attempt to standardize data from scratch. We wouldn't be confident in the output. (2 Points) We could export data from our central system, but it would need significant manual cleaning, re-formatting, and interpretation by an expert before it would be usable. (3 Points) We have established data pipelines and clear governance. We could generate a structured, reliable dataset for a specific use case (e.g., all near-misses from the last 3 years) with relative ease.

7. How integrated is your EHS data with other critical business systems (e.g., HR, Maintenance, Operations)?

(1 Point) Completely siloed. Our EHS data lives in its own system (or spreadsheets) and has no connection to other departmental data. (2 Points) Manual integration. We manually export and combine data from different systems when needed for analysis. (3 Points) Strategic integration. Our EHS system talks to HR, Operations, and Maintenance systems (via APIs or a data warehouse), giving us a holistic view of risk.

8. How does your organization typically approach the procurement of a major software system like an EHS platform?

(1 Point) We issue a detailed Request for Proposal (RFP) with a fixed scope, timeline, and budget, expecting a fully "finished" product on a specific delivery date. (2 Points) We use a traditional project-based approach but build in some room for change requests or a potential "Phase 2" for future enhancements. (3 Points) We look for a long-term strategic partner. Our procurement process is designed to support an agile, iterative collaboration where new capabilities are developed and improved over time.

9. In your view, when would an AI implementation be considered "complete" or "successful"?

(1 Point) When the feature described in the contract is delivered and switched on. (2 Points) After it passes User Acceptance Testing (UAT) and has been running without major bugs for a few months. (3 Points) We don't see it as ever being "complete." Success is measured by ongoing performance metrics and user feedback. We understand the model will need continuous monitoring and retraining to stay relevant.

10. What is your long-term plan for the AI model's maintenance and performance after the initial launch?

(1 Point) We assume the vendor is responsible for making sure it keeps working as part of a standard support agreement. We don't have a specific internal plan. (2 Points) We have a maintenance contract for technical support and bug fixes, but have not yet planned for the processes of model retraining or monitoring for accuracy degradation. (3 Points) Our strategic plan includes dedicated processes for user feedback loops, regular model performance reporting, and scheduled retraining cycles to adapt to new operational realities or regulations.

11. What is the EHS reporting culture in your organization?

(1 Point) Punitive. Incidents (especially near-misses) are often under-reported or hidden to avoid "getting in trouble." We don't have honest data to feed an AI. (2 Points) Neutral/Reactive. Reporting is encouraged, but employees are skeptical that management might use the data against them. Data quality is questionable. (3 Points) Proactive and non-punitive. Leadership actively promotes near-miss reporting as a "learning opportunity." We have a rich, honest stream of data for an AI to learn from.

12. What is the level of commitment for your EHS expert team to validate and train an AI model (Human-in-the-Loop)?

(1 Point) It's seen as "extra work." Our experts are already overloaded; we expect the AI to save them time, not take their time. (2 Points) We can commit time at the beginning (for a PoC), but we don't expect this to be a long-term, ongoing time investment. (3 Points) We understand that "growing" the AI model depends on continuous feedback from our experts. This validation and training process has been formalized as part of their core responsibilities.