Definition
PHA Safeguard Implementability Rate, or PSIR, is the percentage of AI-recommended PHA or HAZOP safeguards that are implementable at the specific facility, meaning the required equipment exists, the procedures are compatible with current operations, and the personnel have the required training. The KPI exists to answer one decision: are the AI-generated safeguard recommendations producing PHA records that look complete but contain safeguards that cannot actually be implemented, leaving hazard scenarios without effective protection?
The Measurement Gap
Adaeze is the process safety engineer for a midstream gas processing site that recently completed a full revalidation PHA on its amine treating unit. The plant''s consulting firm had brought an AI-augmented HAZOP toolkit to the session, a system that ingests the unit''s P&IDs, operating envelope, and a structured deviation taxonomy, then proposes hazard scenarios and recommended safeguards for the HAZOP team to consider. The toolkit had been used successfully at several peer sites and was sold as a way to make PHA sessions more thorough and more consistent.
The PHA produced a fat record. Where prior revalidations had typically generated forty to sixty recommendations for the close-out tracker, this one generated almost three hundred. The session team had felt confident throughout: the AI was raising scenarios the human facilitator might have missed, and the recommended safeguards looked rigorous. The PHA report went to corporate and the close-out process began.
The close-out process is where Adaeze''s problem started. Of the three hundred recommendations, about forty were for new instrumented protective layers (pressure transmitters, temperature elements, gas detectors) that did not exist on the unit. Twenty-eight were for procedural changes requiring operators to follow steps that referenced equipment the unit did not have. Nineteen were for safeguards that called for personnel with specific certifications the site did not have on its current operator and technician roster. A handful referenced control system features that the unit''s aging DCS did not support. And about thirty were safeguards that would technically work but required process modifications that conflicted with the operating envelope corporate had recently locked down through a separate management-of-change process.
Adaeze''s tracker had two hundred and thirty recommendations from this PHA that needed either implementation or formal documented non-implementation. The implementable ones (about a hundred and ten) were being worked through normally. The not-implementable ones, the ones requiring modifications or new equipment, were piling up against a calendar that did not have CAPEX or schedule. The hazards those safeguards were meant to address were real. The PHA had identified them properly. The safeguards in the record were not the right safeguards for this specific unit, this specific configuration, this specific operator roster. They were generic best-practice safeguards that the AI had pulled from its training corpus, where the corpus reflected peer sites'' configurations and not Adaeze''s.
The auditable record now showed that for several serious hazard scenarios, the documented safeguard could not actually be implemented. The hazard had been identified. The protection had been proposed. The protection did not exist. The audit posture was worse than before the AI-augmented PHA because the prior PHA''s recommendations had been chosen by people who knew the unit, and the recommendations they made were the recommendations they intended to implement. The AI-augmented PHA had created a documentation pattern where identified hazards lacked effective protection and the gap was on the record.
Legacy PHA metrics do not catch this. PHA completion rate is good: the session completed on schedule. Recommendation volume is up: the AI is finding more hazards. Time to close-out is the metric that will eventually flag the problem, months from now, as recommendations that should have closed in ninety days are still open at six months and a year. By the time the metric flags the issue, the audit posture has been compromised for a long time.
What is missing is a metric that asks, at the moment the PHA closes, whether the recommended safeguards can actually be implemented at this specific facility.
Formula
PHA Safeguard Implementability Rate, or PSIR, is the percentage of AI-recommended PHA or HAZOP safeguards that are implementable at the specific facility, meaning the required equipment exists, the procedures are compatible with current operations, and the personnel have the required training. The KPI exists to answer one decision: are the AI-generated safeguard recommendations producing PHA records that look complete but contain safeguards that cannot actually be implemented, leaving hazard scenarios without effective protection?
The structural form is a categorization-and-counting exercise applied to each safeguard the AI recommended. A process safety engineer, working with the facility''s equipment and procedure data, scores each recommended safeguard into one of three categories. Implementable means the safeguard can be implemented as recommended: the equipment exists in the right service, the procedures are compatible, and the personnel have the necessary qualifications. Implementable with modification means the safeguard is feasible at the facility but requires defined changes (equipment relocation, procedure revision, training delivery) to make it workable. Not implementable means the safeguard cannot be implemented at the facility as it currently exists, either because the required equipment does not exist and cannot reasonably be added, because procedures fundamentally incompatible with current operations would be needed, or because capabilities required for the safeguard simply do not exist at the site.
PSIR is the percentage of recommendations in the first two categories: implementable as-is or implementable with modification. Direct implementability is a tighter metric that captures only the as-is category. The gap rate is the percentage in the not-implementable category. The three numbers together tell a story that any single number cannot. A facility with high PSIR but low direct implementability has a PHA program that produces useful recommendations but ones that require substantial follow-on work to operationalize. A facility with low PSIR has recommendations the program cannot translate into protection.
Proposed formula, pending validation pilots.