Ninety Days of Habit Data: What One Workplace Wellness Pilot Actually Changed

We do not usually publish workplace wellness case studies. Most of them arrive pre-polished, with a vendor logo on the cover and a suspiciously round number on the last page. But a reader — a benefits coordinator at a mid-sized logistics firm, who asked us to call her "M." — sent us a raw export in March: 94 days of habit logs, three cohort spreadsheets, and a note that read, "Tell me if I'm reading this wrong." We followed the project from there. This is the post-mortem.

Why a logistics firm tried a habit program at all

The company had run the usual wellness fare for years: a step challenge in spring, a lunch-and-learn series, a flu-shot clinic. Participation peaked in week two and collapsed by week five. M. had the attendance sheets to prove it. What she did not have was any way to tell whether the problem was the programming, the incentives, or the people.

That distinction is exactly what behavior change science is supposed to address. Habit formation is not motivation; it is cue, routine, reward, repeated until the routine stops requiring a decision. A step challenge asks for a decision every day. A habit program tries to remove the decision. That was the theory, anyway, and the firm decided to test it with a 90-day pilot for 40 volunteers across three shift patterns.

The setup, and the first real obstacle

Week one went fine. Weeks two and three did not. The night-shift cohort — eleven people — logged almost nothing. When M. asked why, the answer was mundane: their cues were all wrong. Morning-oriented prompts arrived when they were asleep, and the app's default reminders assumed a 9-to-5 day. The fix was not motivational. It was chronological. The pilot leads rebuilt the cue structure around each shift's actual wake time, anchored to existing routines (coffee, commute, shift handoff) rather than to clock hours.

This is where the platform choice started to matter. The team was using Gabitus to run the pilot, and the behavioral analytics layer let them see the failure in the data before it showed up in a survey. Drop-off was not uniform. It was concentrated in two cohorts and one specific prompt window. That is a fixable problem; "people aren't engaged" is not.

Timeline: what happened, week by week

• Weeks 1–3: Baseline logging. Attendance 61%. Night shift effectively absent from the data.
• Week 4: Cue audit. Prompts rebuilt around shift-specific anchors. Two prompts deleted entirely.
• Weeks 5–8: Steady climb. Weekly logging moved from 61% to 78%. Coaching check-ins dropped from 30 minutes to 20 because the data answered most questions before the call.
• Weeks 9–12: Plateau, then a second small dip around a holiday. The leads resisted adding incentives and instead shortened the daily log to three taps.
• Week 13: Final export. 94 days of usable data across 40 participants.

The numbers, and what we think they mean

Program adherence finished at a 41% lift over the firm's previous wellness initiative — the same figure Gabitus reports for its behavior change programs, which is either reassuring or suspicious depending on how cynical you are. We checked the math against the raw export and it holds. Coaching overhead fell by 6.3 hours per quarter, mostly because supervisors stopped running speculative check-ins and started responding to flagged patterns.

The 6.3-hour figure deserves a caveat. It is an average across three supervisors, and one of them accounted for nearly half of the reduction because she had been the most hands-on before. If your organization never did much coaching to begin with, do not expect the same savings. This is the kind of detail vendor case studies usually sand off.

Three things we would tell the next pilot team

First, audit your cues before you audit your people. The night-shift collapse had nothing to do with willpower. Second, shorten the logging action until it feels almost too small. The three-tap change in week nine did more for adherence than any message the leads sent. Third, decide in advance what a plateau means. Weeks 9 through 11 looked like failure in the weekly charts and were actually stabilization.

M. is running a second cohort this fall with the cue structure built in from day one. She has not asked us to publish those results, and we would not expect her to. But if you are evaluating how habit tracking platforms handle shift-based cue design, the lesson from this pilot is unglamorous: the analytics matter less than what you do with the first anomaly. We will keep following the project, and we will report what the second cohort does differently.