Explained

Can AI run a HAZOP? No, and it should not.

The judgment of the people in the room is the study. The honest question is what AI can do to the weeks of preparation and follow-up around that room.

What a HAZOP actually is

HAZOP (n.): a hazard and operability study. A structured team review in which a facilitator walks the design node by node, applying guidewords (no flow, more flow, reverse flow, more temperature) to surface credible deviations, their causes, their consequences, and the safeguards against them. The method is standardized in IEC 61882.

Notice what the definition describes: a team, in a room, thinking. The guidewords are not a formula that produces hazards. They are prompts aimed at human memory and imagination: the operator who remembers what the unit did during the 2019 turnaround, the engineer who has seen this pump configuration fail somewhere else. The deviations that matter most are the ones only experience can rate as credible.

The worksheet that comes out of the room is a record of the study. It is not the study. The study is the collective, accountable judgment of people who will live with the consequences.

Why “AI runs the study” is the wrong goal

A model can generate deviations. It can generate a great many of them, fluently. That is precisely the problem: a HAZOP's value is not the length of its deviation list but the quality of the dismissals. Every deviation the team sets aside as not credible is a judgment someone is accountable for. An AI can propose; it cannot take responsibility for what it dismisses.

The output is defensible judgment, not a filled worksheet. Findings carry the sign-off of people who own the outcome.
The process is the point. The argument in the room, the operator correcting the engineer, is where hazards get found. Automating the room deletes the mechanism.
Regulators and insurers agree. The study exists to put experienced human judgment on the record. A study without it is paperwork.

The honest role

What AI legitimately does to a HAZOP.

Everything around the judgment. A HAZOP consumes weeks, and most of those weeks are not spent judging. They are spent assembling drawings, chasing data, and writing things down.

The machinePrepares and records

Prepares nodes on the as-built plant, from the compiled drawing set, instead of archived drawings of uncertain revision
Pre-screens deviations with evidence: for each guideword, what the topology and operating history actually show
Keeps the record as the session runs, cited to sheets and data
Tracks actions to closure after the room empties, where studies traditionally leak value

The teamJudges and signs

Rates which deviations are credible, from experience no dataset holds
Argues, corrects, and imagines: the actual mechanism of the study
Decides safeguards and actions
Signs the study and owns it

The result is calendar compression, not judgment replacement. The team walks into the room with current drawings, evidence attached to every node, and a scribe that never falls behind. The sessions spend their hours on the deviations that deserve them. Nothing about the judgment changed hands.

How Intuigents prepare a HAZOP

The claim

A vendor claiming AI runs the study misunderstands what the study is for.

A HAZOP exists to put experienced judgment on the record, with names attached. Automate the preparation, the evidence, and the record: all of it, aggressively. Never the judgment. Any pitch that blurs that line should end the meeting.

See the preparation, not a promise.

Book a demo The HAZOP workflow