How to Build a Safety Metric Dictionary in 30 Days

A safety dashboard can look precise while every department is counting a different thing. One plant counts near misses only when a form is opened. Another counts verbal reports. A contractor reports observation volume as proof of field presence, although nobody has defined what qualifies as an observation. The result is not a data problem only. It is a leadership problem disguised as measurement.

The thesis of this guide is practical: a safety metric dictionary should be built before the dashboard is trusted. Without definitions, owners, formulas and review rules, metrics become decoration. Leaders compare numbers that were not produced in the same way, supervisors learn to manage appearance, and EHS spends meetings explaining why last month's curve does not match what the field actually saw.

Across 25+ years in executive EHS roles and more than 250 cultural transformation projects, Andreza Araujo has seen that weak measurement often protects weak decisions. As she argues in Safety Culture: From Theory to Practice, culture appears in repeated choices. A metric dictionary makes one of those choices visible because it forces the organization to say what it means before it rewards, escalates or ignores a number.

What you need before starting the metric dictionary

Before the first workshop, gather the current dashboard, incident classification rules, observation cards, audit forms, contractor reports, absenteeism fields, corrective action tracker and any executive scorecard that uses safety numbers. Add the current owner of each source, even when ownership is informal. A metric cannot be governed if nobody knows where it starts.

Choose one scope for the first 30 days. A plant, business unit or contractor portfolio is better than the whole company, because the goal is to test definitions against real records. If the organization already has a new EHS data analyst role, this dictionary becomes that person's operating map rather than another spreadsheet.

Step 1: List every metric leaders already use

Start with the metrics that influence meetings, bonuses, contractor ranking, board reports and supervisor conversations. Include TRIR, LTIFR, severity rate, DART rate, near-miss volume, observation quality, audit closure, corrective action closure, critical control verification, training completion, absenteeism, psychosocial risk indicators and any local index that appears in monthly reviews.

Do not begin by judging whether each metric is useful. First, make the measurement landscape visible. Many organizations discover that the same word appears in five reports with five meanings. "Closed action" may mean assigned, completed, verified, approved or expired without escalation. "Near miss" may include hazards in one site and exclude them in another.

The verification question is simple enough for the first week. Could a new EHS manager read the list and know which numbers currently shape decisions? If not, the organization is already measuring by memory and habit.

Step 2: Write the operational definition for each metric

For each metric, write what is included, what is excluded and the boundary condition that decides doubtful cases. A near miss should not depend on who tells the story. A recordable injury should not change because the case is politically uncomfortable. A completed corrective action should not mean "uploaded evidence" when the real intent is risk reduction.

This step is where the dictionary becomes cultural work. Andreza Araujo often distinguishes conformity from control, especially in A Ilusao da Conformidade, glossed as The Illusion of Compliance. A metric can comply with a reporting routine while still failing to describe the condition leaders think they are managing.

Use named anchors when the metric depends on a formal source. OSHA recordkeeping rules, the ILO guidance on occupational injury statistics and company incident classification standards can all shape definitions. The local dictionary should state which source applies, because mixed sources create numbers that look comparable but are not.

Step 3: Define the formula and calculation window

Write the exact numerator, denominator, multiplier, period and reset rule. For severity rate, state whether lost days, restricted days or calendar days are counted and which population is included. For observation quality, define whether the score comes from risk relevance, conversation quality, closure evidence or another criterion.

Formula discipline prevents one of the quietest dashboard failures. The same metric may improve because the numerator fell, the denominator rose, the period changed or a late case was moved into a different month. Leaders then interpret movement as performance, although the movement came from arithmetic.

Connect this step with the existing severity rate calculation logic if your dashboard uses that indicator. The dictionary should make clear which decisions the formula can support and which decisions it cannot support alone.

Step 4: Name the data source and evidence owner

Every metric needs a source system, source document, extraction owner and evidence owner. The person who exports data may not be the person who owns evidence quality. For example, EHS may export corrective action status, while maintenance owns the proof that a guard was restored, tested and still works under production conditions.

This distinction matters because weak evidence can travel through a dashboard with impressive speed. A closed action, reported by someone who did not inspect the field condition, can produce a green indicator while the risk remains active. When the dictionary names the evidence owner, it becomes harder to hide behind administrative closure.

The same logic applies to contractor data. If contractors report hours, observations, incidents and action status, the dictionary should state who validates those numbers before they enter owner dashboards. Otherwise, contractor comparisons reward reporting style as much as performance.

Step 5: Add interpretation limits before the metric is used

For each metric, write what the number can suggest and what it cannot prove. TRIR can show reported recordable injury frequency, although it cannot prove fatal-risk control. Training completion can show attendance, although it cannot prove field competence. Near-miss volume can show reporting activity, although it cannot prove learning quality.

This step protects leaders from false confidence. A dashboard without interpretation limits invites people to treat every green cell as safety. James Reason's work on latent failures helps explain why that is dangerous. Harm can be prepared by decisions, conditions and weak barriers that remain invisible until the organization asks questions beyond the injury count.

Use the dictionary to connect each metric with at least one companion indicator. The article on leading indicators TRIR will never show is useful here because it separates injury outcomes from signals that reveal whether controls are present, verified and trusted.

Step 6: Set review frequency and change control

Define how often each metric is reviewed, who can change the definition, and how historical data is handled after a definition changes. A metric dictionary loses credibility when definitions shift quietly after a leadership challenge, acquisition, software migration or contractor dispute.

For stable executive indicators, quarterly review may be enough. For new leading indicators, monthly review is usually better during the first cycle because teams need to see whether the measure produces usable decisions. The important point is not the frequency itself, but the discipline that prevents silent edits.

Create a change log with the date, reason, approver and impact on trend interpretation. If a metric changed in March, the dashboard should not pretend January through June are a single uninterrupted series unless the data was recalculated under one rule.

Step 7: Test the dictionary against ten real records

Before publishing the dictionary, test each high-value metric against ten real records. Use incident files, observation forms, audit findings, action closures, contractor submissions and verification records. Ask two people to classify the same item using the written definition. If they reach different answers, the definition still depends on personal judgment.

This test reveals whether the dictionary is usable under normal pressure. A definition that works only when the EHS manager explains it is not ready for supervisors, analysts or contractors. The wording should be strong enough that a new person can apply it without a side conversation.

Pay special attention to observation quality, because that metric is often corrupted by volume pressure. The article on observation quality blind spots shows why a high count can coexist with shallow risk detection.

Step 8: Publish one-page metric cards for decision meetings

Turn the dictionary into one-page cards for the metrics leaders actually discuss. Each card should include the definition, formula, source, owner, review frequency, interpretation limits, companion indicators and escalation rule. Keep the full dictionary available, but do not expect executives or supervisors to read twenty tabs before a monthly meeting.

The card format changes the conversation. Instead of asking whether the number is good or bad, leaders can ask whether the metric was produced correctly, whether its limit has been respected, and whether the companion indicator confirms the same story. That is how a dashboard begins to govern decisions rather than decorate them.

Andreza Araujo's Safety School and ACS Global Ventures use this kind of operational clarity when helping organizations connect culture, indicators and field controls. The aim is not a prettier dashboard. The aim is a measurement system that makes weak risk decisions harder to hide.

Final safety metric dictionary checklist

Use this checklist before releasing the first version of the dictionary:

• Every dashboard metric has an operational definition with inclusions and exclusions.
• Each formula states numerator, denominator, multiplier, period and reset rule.
• Source system, extraction owner and evidence owner are named separately.
• Interpretation limits are written before leaders use the metric in decisions.
• Review frequency, approver and definition-change log are documented.
• High-value metrics were tested against real records by more than one person.
• Decision meetings use one-page metric cards instead of unexplained dashboard cells.

A safety metric dictionary is not administrative housekeeping. It is a control over the way leaders see risk. When the organization defines numbers carefully, it reduces the space for cosmetic performance and gives EHS a stronger basis for challenging decisions that look safe only because the dashboard never learned to ask the right question.