How to Do a Safety Risk Assessment

A safety risk assessment identifies the hazards in a task or area, scores each by how likely harm is and how severe it would be, ranks them on a risk matrix, drives each one down the hierarchy of controls, and then re-scores the residual risk that remains. It is what turns a list of hazards into a ranked plan you can act on.

Every plant already knows most of its hazards. What it usually lacks is a defensible way to say which ones to fix first, how far to go, and whether the fix was enough. That is the job of a risk assessment: it takes the gut feeling that "that press worries me" and turns it into a number you can compare, act on, and revisit. It is the core requirement of ISO 45001 clause 6.1.2 and the foundation of machinery safety. This guide walks the full method, identify, score, rank, control, re-score, with a worked example.

What is a safety risk assessment?

A safety risk assessment is the combined process of finding hazards and evaluating the risk each one presents, so controls can be prioritized. Hazard identification asks "what here can cause harm?" Risk assessment asks "how bad, and how likely?" and produces a ranking. The two together are often called HIRA, hazard identification and risk assessment. It is a close relative of the job safety analysis: a JSA breaks one job into steps and controls each step's hazard, while a risk assessment can cover a task, a machine, an area, or a change, and adds explicit scoring so competing risks can be compared. ISO 45001 requires the process to be ongoing and proactive, done before harm, and repeated when things change, not a one-time binder.

One judgment matters before you score anything: assess the risk as the work is really done, not as the procedure says it should be. If operators routinely reach into a machine because the official method is too slow, the honest hazard is the reach, not the tidy version in the manual. A risk assessment built on the paper process instead of the real one produces low scores and real injuries. This is why the people who do the job have to be in the room, they know which corners get cut and why, and that knowledge is the difference between an assessment that protects people and one that protects the file.

How do you score a risk: likelihood times severity

Risk is scored as a combination of two things: how likely the harm is to occur, and how severe it would be if it did. Most plants use a simple scale for each, say 1 to 5 for likelihood (rare to almost certain) and 1 to 5 for severity (first aid to fatality), and multiply them for a risk score from 1 to 25. That score is what lets you compare a frequent minor cut against a rare crushing injury on the same page. The matrix below is the tool that does it: it plots likelihood against severity and colors the cells so a glance tells you what is intolerable, what needs work, and what is broadly acceptable.

How do you drive risk down with the hierarchy of controls?

Once a risk is scored, you reduce it by working down the hierarchy of controls in order, not by picking whichever control is cheapest. ISO 45001 is explicit that the hierarchy is a sequence, not a menu: eliminate the hazard first, then substitute something less hazardous, then use engineering controls, then administrative controls, and only then rely on PPE. The reason for the order is reliability. Elimination and engineering controls work whether or not a person remembers a rule; administrative controls and PPE depend on human behavior every single time, so they sit at the bottom. A guard on a machine, see machine guarding beats a sign that says "keep hands clear," because the guard does not depend on anyone reading it.

What is residual risk, and when do you stop?

Residual risk is the risk that remains after your controls are in place, and you keep reducing it until it is as low as reasonably practicable and inside your acceptance criteria. Every control reduces the score; you re-score after each one. The stopping point is not zero, most real risks cannot be driven to zero, but the level your organization has defined as acceptable, given its legal duties and context. This is often called ALARP: as low as reasonably practicable. The judgment is whether the effort and cost of the next control are grossly disproportionate to the further risk reduction it would buy. Document that judgment; "we accepted a residual score of 4 because further reduction required rebuilding the line for negligible benefit" is a defensible record. Silence is not.

A worked scoring example

Take a shared task: an operator manually clearing jams from a moving conveyor. Here is how the scoring and control might run.

Notice the severity barely moved, a caught hand is serious no matter what, while the likelihood fell sharply. That is typical: controls usually cut how often harm happens more than how bad it would be, which is exactly why eliminating exposure ranks above merely reducing it.

What do the standards require?

The requirement to assess risk is written into the recognized frameworks, and the numbers behind it are public.

• ISO 45001:2018 clause 6.1.2 requires an ongoing, proactive process to identify hazards and assess occupational health and safety risks, and to apply the hierarchy of controls to reduce them.
• OSHA's hazard identification and assessment guidance makes hazard identification and risk assessment a core element of a safety and health program, and stresses finding and fixing hazards before they cause harm.
• U.S. private industry recorded about 2.6 million nonfatal injuries and illnesses in 2023 roughly 946,500 with days away from work (BLS), the harm a good assessment is meant to head off before it happens.

How do you actually run one? A seven-step method

This is the full loop, and it is the same whether you are assessing a task, a machine, or a change.

1. Define the scope. Name exactly what you are assessing, this task, this machine, this area, this change, so the hazards you find are the ones that belong to it.
2. Identify the hazards. Walk it, talk to the people who do it, and review incident and near-miss history. List every way it can cause harm, including non-routine and maintenance conditions.
3. Assess likelihood and severity. For each hazard, judge how likely harm is and how bad it would be, using your defined scales and the existing controls already in place.
4. Score and rank on the matrix. Multiply likelihood by severity, plot it, and sort by score so the intolerable risks rise to the top of the action list.
5. Apply the hierarchy of controls. For the highest risks, work down from elimination, choosing the most reliable control you can practically install.
6. Re-score the residual risk. Score again with the new controls in place, and keep going until the residual risk is as low as reasonably practicable and acceptable. Record the acceptance decision.
7. Record, act, and review. Document the assessment, assign owners and dates to the controls, and revisit it on a schedule and whenever the work, equipment, or people change.

Why do risk assessments end up on a shelf?

The most common failure is not bad scoring; it is that the assessment gets written once, filed, and never revisited while the controls it assumes quietly rot. The interlock that scored the conveyor down to a 4 gets bypassed, the procedure drifts, the line changes, and the paper still says 4. A risk assessment is only true as long as its controls are real, which means it has to be a living record tied to the plant, not a document. That is where the assessment connects to the wider safety management system: the controls it names need owners, due dates, and verification, and the whole thing needs to resurface when something changes. Plants that keep assessments and their control checks visible, the same guard and interlock checks captured where the work happens rather than in a binder, keep the score honest. See how Harmony keeps hazards, controls, and follow-up in one place on the feature overview.