How to Build a Safety Metric Data Dictionary for Monthly Reviews in 14 Days

A monthly safety review can contain twenty numbers and still leave leaders unable to answer one practical question: what should change before the next meeting? The problem is often not a missing dashboard. It is an undefined metric that has no stable scope, owner, or decision attached to it.

This guide shows EHS managers and operations leaders how to build a safety metric data dictionary in 14 days. The purpose is not to document every field in the software. It is to make the few measures used in monthly decisions precise enough that different sites, managers, and analysts interpret them in the same way.

A safety metric data dictionary is a controlled record that explains what a measure means, how it is calculated, who owns the data, how often it is reviewed, and what decision follows when the signal changes.

What do you need before starting?

Choose one monthly review, one accountable sponsor, and no more than twelve measures for the first cycle. Include the EHS lead, an operations representative, a data analyst, and one manager who must act on the information. If the review has no decision owner, the dictionary will become a filing exercise.

Start with the questions leaders already ask. They may want to know whether a critical control was verified, whether overdue actions are accumulating, whether exposure is changing, or whether a serious concern is reaching the right level. Each question should produce a definition that can survive a handoff between departments.

Step 1: List the decisions the review must support

Write the decisions before listing the measures. For example, a plant manager may need to decide whether to stop a recurring task, release capital for an engineering control, add supervision to a shift, or escalate an unresolved action to the executive team.

For each decision, record the person who can make it and the evidence that person needs. This prevents a common failure in which the organization chooses a metric because it is easy to export, then searches for a use after the dashboard is built.

Step 2: Group measures by what they actually show

Place each candidate measure into one of four groups: activity, exposure, control verification, or outcome. The grouping does not make one class superior to another. It makes the limits of each signal visible.

This classification is one of the article's central safeguards. A completed training count and a verified critical control do not answer the same question, even when both appear under a heading called leading indicators.

Step 3: Write the plain-language definition

Give each measure a sentence that a supervisor can understand without opening the reporting system. State what is counted, what is excluded, and the period covered. Avoid definitions such as “overall safety performance,” because they describe an aspiration rather than a measurement rule.

Use a fixed pattern. “This measure counts...” should be followed by the unit, scope, time window, and condition that qualifies a record. If the definition cannot be read aloud without qualification, the measure is not ready for executive use.

Step 4: Document the formula and counting boundary

Record the formula, denominator where one exists, rounding rule, treatment of duplicates, and the point at which an item enters or leaves the reporting period. A number can change because the work changed, because the data rule changed, or because a late record was added. The dictionary must help the reviewer tell those explanations apart.

Do not hide judgment inside a spreadsheet formula. If a supervisor decides whether two reports describe one event or two, document that decision rule and name the role that resolves disagreement.

Step 5: Trace the source and owner

Name the source system, table or form, extraction date, validation check, and data owner. Then name the decision owner separately. The person who maintains a record may not have authority to change the work, and confusing those roles creates silent gaps.

Across 25+ years of EHS leadership, Andreza Araujo has consistently connected safety culture with the decisions that follow evidence. That principle is reflected in Safety Culture: From Theory to Practice, where a measure has value only when it supports accountable action rather than reporting theater.

Step 6: Set the review cadence and trigger

Define when the metric is reviewed and what movement requires attention. A monthly cadence may be appropriate for executive governance, while a shift-level control measure may need a faster response. The dictionary should preserve both the formal review date and the immediate escalation rule.

Triggers can be threshold-based, trend-based, or condition-based. A condition-based trigger may apply when a required control is unavailable, even if the dashboard number remains inside its target range. This is why a green status cannot replace a clear decision rule.

Step 7: Test the definition against real records

Use two completed reporting periods and ask different people to calculate the result independently. Compare their interpretations before comparing their arithmetic. Disagreement about scope is a definition problem, while disagreement about arithmetic is a process problem.

Test an ordinary month, a month with a late report, and a month with an unusual operational change. If the number changes meaning whenever conditions become difficult, revise the dictionary before publishing the measure to leaders.

Step 8: Approve the dictionary and retire ambiguous measures

Ask the sponsor to approve the definition, owner, trigger, and review forum together. Do not approve a metric because it has a polished visualization. Approve it because the organization can explain what it means and what it changes.

Retire measures that have no owner, no decision, no reliable source, or no credible interpretation. Removing an attractive but ambiguous number is a governance improvement, not a reporting failure.

How should leaders use the dictionary in the monthly review?

Keep the dictionary beside the dashboard and use it when a number moves. Ask four questions in sequence. What changed in the work or in the data rule? Which control or exposure does the signal represent? Who owns the next decision? By when will the decision be verified?

That sequence keeps the review from turning into a performance ceremony. It also protects teams from being judged against numbers whose scope was never agreed. The strongest metric review is not the one with the most indicators. It is the one that makes the next control decision easier to see.

What should the 14-day delivery plan look like?

• Days 1 to 2: Select the review sponsor, decision questions, and initial measures.
• Days 3 to 5: Classify measures and write plain-language definitions.
• Days 6 to 8: Document formulas, sources, owners, and counting boundaries.
• Days 9 to 11: Set triggers and test definitions against two reporting periods.
• Days 12 to 14: Resolve disagreements, approve the dictionary, and retire measures that cannot support a decision.

A dashboard can wait for its next redesign. An undefined metric should not wait for its next serious decision. Clarify the measure before the organization uses it to accept risk, fund a control, or declare improvement.

Build the dictionary around the decisions that protect people, not around the fields that are easiest to export. When the definition, owner, trigger, and action stay connected, the monthly review becomes a management control instead of a presentation about numbers.

For more conversations about leadership, safety culture, and the decisions behind prevention, explore the Headline Podcast and Andreza Araujo's books and safety resources.