How Cam Stevens Thinks About Voice Data Before It Becomes a Safety Metric

Episode 15 of Headline Podcast was published on March 18, 2026, and features Cam Stevens, CEO of PKG, in conversation with Andreza Araujo and Dr. Megan Tranter. Stevens defended a practical thesis: voice technology should be judged by the safety decision it improves, not by the amount of data it gathers.

Why should voice data be treated as a safety decision?

Voice data belongs in safety measurement only when it answers a defined operational question. A comment count, sentiment score or automated transcript may look informative, yet none of those outputs proves that exposure fell or that a control became stronger.

Stevens put the sequence plainly on Headline Podcast. “If we're very clear on the problem to solve, then a technology catalog is excellent. If we're not clear on the problem to solve, a technology catalog is very dangerous.” The warning applies to metrics as much as it applies to devices, because a dashboard can give an undefined problem the appearance of precision.

That distinction matters for EHS leaders who are asked to demonstrate progress every month. A new tool can produce thousands of entries while leaving the decision that matters untouched. The first test is therefore not whether the platform works. It is whether the organization can name the risk, the decision and the person who must act.

What does a useful voice signal actually measure?

A useful voice signal measures a change in information quality, decision timing or control reliability. It does not measure participation in the abstract.

For example, a maintenance team might use a voice channel to flag recurring isolation problems before a job begins. The meaningful outcome is not 214 recordings. It is whether the recurring defect was verified, corrected and prevented from returning during the next work cycle.

OSHA's worker-participation guidance explains why worker knowledge matters during hazard identification and control selection. The guidance does not turn every comment into a metric. It supports a process in which information reaches the people who can remove or reduce the hazard.

Andreza Araujo's work across more than 250 cultural transformation projects reinforces this point. Participation becomes credible when people can see what changed because they spoke. Without that visible connection, the organization is measuring willingness to report while quietly teaching workers that reporting has no operational consequence.

How can leaders separate signal from dashboard noise?

Leaders separate signal from noise by linking each data point to a decision rule. If the rule is absent, the organization is collecting observations rather than operating an indicator.

A disciplined metric has at least four parts. It defines the exposure or failure condition, identifies the decision owner, sets a response window and specifies how effectiveness will be checked. Those parts convert a voice entry into a controlled management process.

ISO 45001's emphasis on consultation, participation and performance evaluation supports this connection between information and action. The standard is not a reason to create another scorecard. It is a reason to examine whether the management system can turn worker input into improved controls.

One practical test is to sample 12 voice reports at the end of a month and classify each as acknowledged, investigated, controlled or still exposed. The classification reveals more than a total volume figure because it shows where the response chain breaks.

What psychosocial risks can voice technology introduce?

Technology can make speaking easier, but it can also make workers feel observed. The difference depends on purpose, access, feedback and the consequences attached to the data.

Stevens said that “The changing shift in risk profile will be overwhelmingly psychosocial, driven by technology usage in our organizations.” That observation changes the evaluation criteria. A tool that identifies one hazard while increasing uncertainty about monitoring may transfer risk from the physical system into the work design.

The ILO's guidance on psychosocial risks connects work organization, workload, autonomy and social relationships to worker health. The WHO's mental health at work guidance likewise treats organizational conditions as part of prevention. A voice channel should therefore be reviewed for pace, autonomy and trust, not only for technical accuracy.

In projects shaped by Andreza Araujo's safety-culture work, a leader should ask whether the new channel makes it safer to speak or simply easier to record people. That question is especially important when voice data is linked to names, location, shift performance or individual productivity.

How should a metric protect worker voice?

Worker voice is protected when the organization explains what is collected, limits access, separates learning from punishment and shows how concerns are resolved.

The safeguards should be designed before the pilot begins. Write a short data-use statement, name the decision owner, define the retention period and identify the route for challenging a harmful interpretation. Workers should also know whether the tool accepts anonymous information and what happens when a report indicates immediate danger.

These controls are not administrative decoration. They determine whether the signal remains credible after the first uncomfortable report. A channel that captures criticism but sends it into a closed dashboard will eventually produce safer-sounding language and weaker information.

Andreza Araujo's book Safety Culture: From Theory to Practice is useful here because it treats culture as something visible in decisions and routines, not as a slogan separated from work. The metric should therefore be judged by the routine it creates after a worker speaks.

How does voice data compare with traditional safety indicators?

Voice data adds immediacy and context, while traditional indicators provide structure and trend visibility. Neither source is sufficient on its own.

The comparison shows why a voice channel should complement, rather than replace, control verification. A rising report volume may mean conditions worsened, trust improved or the tool became easier to use. Only a field check and a decision record can distinguish those possibilities.

The Headline Podcast discussion also supports a wider warning. Cam Stevens said, “We can elevate the human experience with technology, but there are certainly times when technology should absolutely not be used.” The decision not to automate is itself a safety decision when the tool would weaken privacy, autonomy or honest reporting.

What should an EHS leader do in the first 30 days?

The first 30 days should test one problem, one signal and one response route. A broad rollout makes it impossible to tell whether the technology improved a decision or simply multiplied activity.

1. Days 1 to 5: define one exposure, one intended decision and one named owner.
2. Days 6 to 10: agree on access, anonymity, retention, escalation and protection against retaliation.
3. Days 11 to 20: run the channel with one team and compare reports with direct field verification.
4. Days 21 to 30: review 12 cases, classify response quality and ask workers whether the tool changed what leaders did.

Use the results to decide whether the metric deserves a place in the management review. If the answer depends on a higher volume of data rather than a clearer control decision, pause the rollout. A smaller signal with a visible response is more valuable than a large feed that no manager can interpret.

Who owns the metric after the pilot ends?

A voice-data pilot should not become permanent simply because the software is available. Before the review closes, the sponsor should record which decision improved, which exposure remained unresolved and which part of the process must change before the next cycle.

Ownership also needs a boundary. The EHS team may coordinate the signal, but the line manager, maintenance lead or project owner must own the operational response. When every issue is routed back to EHS, the metric becomes a transfer mechanism rather than a control mechanism.

A monthly management review can keep the measure honest by asking three questions. Did the signal identify a condition that other indicators missed? Did the responsible owner act within the agreed window? Can workers see the result without exposing individual identities? If the answer to any question is unclear, the metric needs redesign before it is compared across sites.

Recommendation

Cam Stevens's Episode 15 offers a disciplined rule for safety technology: define the problem before selecting the tool, and define the response before counting the signal. Safety leaders should pilot voice data only where the information can change a named decision, then verify whether the field condition actually improved.

That approach aligns with the distinction between activity and evidence in metric validity, the need to distinguish closure from verification in safety metrics, and the discipline of calibrating a leading measure before it reaches the board in leading-indicator calibration.

Listen to the full conversation with Cam Stevens on Headline Podcast. The episode's central lesson is simple without being simplistic: technology earns trust when it improves the work decision and leaves people safer to speak.