Most workplace wellness pilots fail to produce a decision. They run, people report liking it, and the organisation is left with the same question it started with: should we buy this?

The cause is almost always design rather than the intervention. A pilot that did not define success before it began cannot demonstrate it afterwards. This is a template for one that can, and it is deliberately vendor-neutral, since the same structure works whatever you are testing.

Before You Start: Is the Problem Structural?

Worth asking honestly, because it determines whether any pilot is worth running. If your stress driver is chronic understaffing, unmanageable workload, or a specific manager, an app will not fix it, and running a wellness pilot in that situation reads to employees as the organisation treating a structural problem as a personal one.

Stress interventions work on the load that remains after the structural issues are addressed. They are not a substitute for addressing them, and a pilot launched into visible organisational dysfunction tends to produce a backlash rather than a result.

Step 1: Choose One Cohort

Pick a single team of 25 or more rather than spreading a small number of licences across the organisation.

Concentration matters for two reasons. Statistically, a scattered pilot produces a sample too small and too heterogeneous to interpret. Socially, adoption in workplace tools is contagious within a team and invisible across teams. Twenty-five people in one department will produce a clearer answer than a hundred spread across eight.

Pick a team with a real stress load, and avoid the team that volunteered enthusiastically. Motivated volunteers produce results you cannot generalise from.

Step 2: Baseline Before Launch

Measure before anyone gets access. This is the step most often skipped and the one that makes results defensible.

A validated instrument. The Perceived Stress Scale is the common choice: short, well-established, free, and comparable to published literature.
A physiological measure, if available. Heart rate variability against each participant's own baseline, never compared between individuals.
Two or three operational metrics you already collect. Absence, voluntary attrition, or a relevant engagement survey item. These are noisy over a quarter but cost nothing to include.

Baseline over at least two weeks. A single measurement day captures whatever was happening that week.

Step 3: Agree the Success Threshold in Advance

Write down, before the pilot begins and with the vendor's agreement, what result would cause you to buy and what result would cause you to walk away.

A workable default: a meaningful reduction in the psychometric measure, plus 90-day engagement above an agreed share of enrolled employees. Fix both numbers in advance. Thresholds set after seeing the data are not thresholds.

This step also functions as vendor selection. A vendor unwilling to agree a threshold in writing is telling you something about their confidence in the outcome.

Step 4: Design for the Working Day

The practice phase is where pilots die, and the failure is mechanical. An intervention that requires finding a quiet room, sitting down, and opening a phone competes with the working day and loses by week three.

Whatever you pilot should fit inside the moment where the stress occurs. In practice that means short, ideally under five minutes, available without a screen, and usable without announcing to colleagues that you are doing it. This constraint eliminates a surprising number of otherwise good products.

Step 5: State the Privacy Model Publicly

Before enrolment opens, tell participants in plain language: what is collected, who sees it, and at what level of aggregation.

The defensible model is that individual data stays with the individual, the employer sees aggregates only, aggregates have a minimum cohort size, and opting out carries no consequence. Say it in the launch email rather than burying it in a policy document. Participation rates respond to this more than to incentives, and a pilot where people suspect surveillance measures suspicion rather than stress.

Step 6: Report Honestly

Report	Not
Engagement at day 30 and day 90, separately, as a share of enrolled employees	Total sessions or minutes
Mean change in the validated instrument, with the number who completed both measurements	Selected testimonials
Share of participants improving against their own baseline	Between-person comparisons or rankings
What did not move	Only the metrics that improved

Report the dropouts. If 40 of 60 enrolled employees completed the pilot, the result describes those 40, and a reader who discovers the attrition later will discount everything else in the report.

A Realistic Timeline

Weeks 1 to 2: baseline measurement, privacy communication, enrolment
Weeks 3 to 10: practice phase, with a light mid-point check on engagement only
Week 11: re-measure with the same instrument
Week 12: report against the pre-agreed threshold and decide

One quarter, one cohort, one decision. That is enough, and pilots that run longer without a threshold usually run until everyone forgets why they started.

Running This With Vayu

We offer a free 30-day pilot for teams of 25 or more, and the structure above is roughly how we would recommend running it regardless of which tool you choose. Vayu delivers haptic-guided breathwork on Apple Watch and Wear OS with HRV-adaptive pacing, sessions average about two minutes, and employer reporting is aggregate-only by design. Team pricing after a pilot runs $4 to $8 per employee per month.

If the pilot does not clear the threshold you set, that is a useful result too, and considerably cheaper to discover in one quarter than in one renewal cycle.

A pilot without a pre-agreed threshold is not a pilot. It is a trial period that ends in an opinion.

Key Points

Confirm the stress driver is not structural before piloting software against it.
Concentrate on one team of 25 or more rather than scattering licences.
Baseline for two weeks with a validated instrument before anyone gets access.
Agree the buy and walk-away thresholds in writing before starting.
Publish the privacy model in plain language at launch.
Report engagement separately at 30 and 90 days, include dropouts, and say what did not move.