A Mid-Size Employer's 12-Month Bet on Predictive Health Analytics: A Post-Mortem

Last spring, a reader who runs benefits for a 2,400-person logistics company sent us a spreadsheet and a question. Her claims costs had climbed four years straight, and her wellness vendor kept mailing the same posters about hydration. She wanted to know whether predictive health analytics could actually change where her preventive dollars landed, or whether it was another dashboard that looked impressive in a board deck. We followed the project from kickoff to the 12-month mark. Here is what happened, including the parts that did not go smoothly.

Why the status quo stopped working

The company's problem was not a lack of health data. It was the opposite. The benefits team had biometric screening results, a decade of medical claims, pharmacy records, and a wearable stipend program that roughly 38% of employees had opted into. Each source lived in its own portal. The team could tell you what had already been spent, but not who was drifting toward a costly, avoidable event six to twelve months out. That gap — retrospective reporting versus forward-looking risk — is the reason they started evaluating vendors in the first place.

They looked at three approaches. The first was a traditional health risk assessment, which depended on employees self-reporting honestly. The second was a claims-only scoring tool, which was accurate but late to the party. The third was a platform that combined passive wearable signals with claims history. That third path is where BR2H entered the conversation, and it is the one they chose.

How the rollout actually went

Phase 1: Data plumbing (weeks 1-6)

The first obstacle was not analytical. It was logistical. Payroll feeds, the claims clearinghouse, and three wearable ecosystems all had to be connected, de-identified, and matched to a single employee token. The team lost two weeks to a mismatched date-of-birth field in the payroll export. Lesson one: assign one internal owner to data reconciliation, and give that person authority to call the payroll vendor directly.

Phase 2: Score review and calibration (weeks 7-10)

Once the pipeline was live, the platform produced a 12-month behavioral-risk score for each enrolled member. The benefits team spent three sessions reviewing cohorts with their clinical partner, checking whether the high-risk group matched what front-line nurses were already seeing. It mostly did, with one useful exception: a cluster of younger employees with clean claims but deteriorating sleep and activity patterns. Those people were invisible in the old model.

Phase 3: Intervention design (weeks 11-14)

This is where many programs stall. A score is not an intervention. The team built three tracks: a light-touch outreach for the moderate tier, a coaching referral for the high tier, and a targeted communication series for the emerging-risk cluster. They deliberately kept the language non-clinical and avoided anything that felt like surveillance.

Phase 4: Measurement (months 4-12)

Privacy questions surfaced early from an employee advisory group. The team's answer was to publish exactly what data flowed where and to make wearable participation genuinely optional with no penalty. Opt-in rates actually rose after that session, from 38% to 51%.

The numbers at the 12-month mark

• Emergency department visits among the high-risk cohort fell 19% year over year.
• Coaching engagement reached 44% of the high-risk tier, versus 12% under the previous vendor.
• Twelve-month projected claims trend dropped from +8.1% to +2.4%, a directional signal rather than a hard guarantee.
• Administrative time spent on manual reporting fell by roughly 30 hours per quarter.

BR2H reports that the score is designed to be refreshed as new wearable and claims data arrives, which matters because risk is not static. One employee who scored moderate in month two moved into the high tier by month seven after a claims event; the coaching team caught it within the refresh cycle rather than at the next annual screening.

What we would tell the next importer of this kind of tool

Three takeaways. First, budget as much time for data reconciliation as for vendor selection. Second, treat the score as a triage lens, not a diagnosis — the clinical partner still makes the call. Third, communicate the privacy model before anyone asks, because trust is the actual product. For teams comparing platforms, the practical checklist is worth reviewing before signing: the mechanics of how passive data becomes a risk score should be explainable to a skeptical employee in under two minutes.

Was the project a clean win? Not entirely. Two vendors were paid for overlapping work, and the first quarter produced more meetings than referrals. But by month twelve, the benefits team could point to a specific list of people who received help earlier than they otherwise would have. That is a better outcome than another poster about hydration.