Job Hazard Analysis: Written at a Desk, Ignored on Site

The job hazard analysis for the task is in a binder in the site office. It was written two years ago by somebody who has never done the task. It lists a machine that was replaced last spring, and under controls, in four places, it says be careful.

What a job hazard analysis is

A job hazard analysis takes one job, breaks it into its steps, asks what could hurt somebody at each step, and records what will be done about it. It is also called a job safety analysis. Three columns do the work: the step, the hazard, the control.

The idea is simple and it works when the document describes the job as it is actually done. That condition is where they tend to fail.

How they go wrong

Failure	How it happens	What it costs
Written at a desk	Somebody who does not do the job writes the steps	It describes the procedure, not the practice
Copied from another job	A template is reused with the title changed	Hazards that are not there, and none of the ones that are
Never updated	New equipment or a new site, same document	Controls for a machine you sold
Vague controls	Be careful, use caution, wear PPE	Nothing a person can do differently
Filed and forgotten	Signed once, kept in a binder	Nobody on the job has read it

Hunt for the words be careful. They turn up in control columns again and again, and they are not a control. A control changes the job: a guard, a lockout, a second person, a different tool, a rule about when the work stops.

Safety practice ranks controls in a widely used order: remove the hazard, substitute something safer, engineer a barrier, change how the work is organised, and protective equipment last. A document where every row ends in protective equipment has usually skipped the harder questions.

Where AI earns its place

A first draft of the steps. Given the work order or the written procedure, it can lay out a step list for the crew to correct. Correcting a draft is faster than starting from a blank page, and the crew's corrections are where the real job shows up.

Bringing your own history to each step. Near misses and incidents from your records, matched to the step they happened on, are the best evidence of what the hazards are. Cross-referencing them by hand rarely happens, and it is the same connection that improves incident reporting.

Flagging stale documents. An analysis that names equipment you no longer own, or has not been reviewed since the process changed, can be found by comparing it with your asset list and your change records.

Hunting for non-controls. Finding every row where the control is a reminder to be careful is a quick scan, and the list it produces is a to-do list for whoever owns safety.

Plain language and translation. A document the crew can read in the language they work in gets used, with a fluent person checking it.

What must stay with people

Whether a control is adequate. That is a professional judgment about a specific task, site and crew, and it carries legal weight. A model will rate a control as sufficient in a confident sentence without having seen the job.

What the law requires. Health and safety requirements in Canada are set by province and territory, with a separate federal regime, and they vary by trade and task. Take them from your regulator, your safety association or a qualified safety professional, and check any draft against them.

The review by the people who do the work. The single most useful step is the crew reading the analysis and saying that is not how we do it. Nothing generated replaces that, and an analysis that skipped it is the desk-written document again.

Keeping it alive

Tie the review to events, not the calendar. New equipment, a new site, a change of materials, an incident or a near miss on that job should each trigger a look. An annual review catches the change eleven months late.

Then use it. The analysis should feed the morning briefing for that job, as in toolbox talks built from your own records, and travel with the work order. The serious hazards it identifies belong on the risk register with a named owner.

Frequently Asked Questions

What is a job hazard analysis?

A document that breaks one job into steps, identifies what could cause harm at each step, and records the control for each hazard. It is also known as a job safety analysis. Its purpose is to make the hazards of a specific task visible to the people doing it before they start.

Why do job hazard analyses fail?

Usually because they do not describe the real job. They are written by someone who does not do the work, copied from another task, left unchanged after equipment or sites change, or filled with vague controls such as be careful. A document like that gets signed and ignored.

What counts as a control?

Something that changes the job. Removing the hazard, substituting a safer material or method, a physical barrier or guard, a change to how the work is organised, and protective equipment are the usual categories, in that order of preference. A reminder to be careful is not a control.

How can AI help with a job hazard analysis?

It can draft the step list from a work order or procedure for the crew to correct, match your own near misses and incidents to the steps where they happened, flag analyses that mention equipment you no longer have, find rows where the control is only a reminder, and produce plain-language or translated versions for checking.

Can AI decide whether a control is adequate?

No. Adequacy is a professional judgment about a specific task, site and crew, and it has legal consequences. Legal requirements also vary by province, territory and trade. Those decisions belong to a qualified safety professional and your regulator, and the crew doing the job must review the result.

How often should a job hazard analysis be reviewed?

Whenever the job changes, and after any incident or near miss on that job. New equipment, a new site, different materials or a different crew are all triggers. A calendar review alone will be late, because it catches a change months after it happened.