Wellness Program KPIs: Participation Is Not Engagement, and Neither Is an Outcome. The Seven Numbers to Report Instead

Every wellness program dashboard leads with the same number: participation. It is usually between 30% and 40%, it is usually presented as a success, and it is the least informative figure on the page.

Three decades of evidence explain why. RAND’s study for the US Department of Labor found that employers with no incentives reported median participation of 20%, employers offering rewards reported 40%, and employers using penalties or surcharges reported 73% [1]. That is not three levels of program quality. It is three sizes of incentive. Participation measures the offer.

Then two randomised trials, the only large ones in the field, showed what participation does not measure. At the University of Illinois, 56% of employees offered the program took part, and the people who took part had lower medical spending and healthier behaviour than those who did not before the program had started [2]. At BJ’s Wholesale, self-reported exercise and weight management improved at 18 months while cholesterol, blood pressure, BMI, spending, utilisation and absenteeism did not move, and the same was true at three years [3][4]. Participation rose. Health did not.

Programs are still funded on the participation number, at a median incentive of $600 per employee in 2025 [5], because nobody has handed the people funding them a better set of numbers. This is an attempt to do that: seven KPIs to sit beside participation, defined, sourced, with the ranges the evidence supports.

The funnel, and where the evidence says it breaks

Each row is discussed below. The important structural point is the order: verification comes before persistence, and selection comes before outcomes. A program that reports outcomes without the two rows above them is reporting who joined.

1. Activation rate

Registered accounts divided by the eligible population. It is the honest top of the funnel and almost never appears in a quarterly report, because it is typically the worst number in it. Report it anyway, because the gap between activation and participation tells you whether the problem is awareness or the product.

2. Participation rate

The number everyone has. Define it precisely: the share of eligible people who completed at least one qualifying action in the period. Then read it as what it is.

RAND’s Department of Labor study, a nationally representative employer survey, gives the cleanest calibration. Median participation was 20% where no incentive was offered, 40% where rewards were offered, and 73% where penalties or surcharges were used [1]. Configuration mattered too: comprehensive programs reported a median of 59%, prevention-focused programs 45%, and intervention-focused programs 28% [1]. The same study’s analysis of a Fortune 100 employer’s seven years of data found that a non-participation surcharge raised smoking-cessation uptake by 8.5 percentage points, to a level still below 30%, and was associated with a 19-point fall in disease-management participation [1]. Penalties buy compliance in some places and drive people away in others.

The trials calibrate the other way. BJ’s, with a program of nutrition, activity and stress modules run by dietitians: 12.2% completed the first module, an average of 30.6% completed later ones, 35.2% of everyone ever employed at treatment sites completed at least one, and 21.4% completed three [3]. Illinois, with a well-resourced university program and cash incentives for completing its components: 56% [2].

So a 35% participation rate with rewards is about the median. A 35% rate with a penalty attached is poor. A 35% rate with no incentive is excellent. The number means nothing without the incentive next to it.

KFF’s 2025 survey shows how common the incentive is: 83% of large firms offer at least one wellness program, 43% offer biometric screening, and among those, 62% use incentives or penalties to get it completed. 53% offer a health risk assessment, and 53% of those attach an incentive [11].

3. Connected-device rate

The share of participants whose behaviour is verified by a data source rather than typed into a form. This is the KPI most programs do not have, and it is the one that determines whether everything below it is measurement or testimony.

Its ceiling is device ownership. Rock Health’s December 2025 survey of 8,000 US adults put wearable ownership at 46%, and described owners as younger, wealthier, more urban, healthier and more likely to be commercially insured than non-owners [6]. A program that verifies only through wearables is therefore capped below half its population and skewed toward the members who were already active. In Britain the ceiling is 35%; in the EU 30%; we compiled the national figures in the adoption reference.

The phone changes the ceiling. 91% of US adults own a smartphone [7], and the phone records steps, movement and a usable sleep estimate without any wearable. UnitedHealthcare Rewards, for one, already accepts “activity tracker, smartwatch or smartphone” as the verifier [12]. A program that counts the phone raises its connected-device rate from the wearable-owning minority to nearly everyone, and, more importantly, stops selecting for the healthy at the verification step.

Report the rate, and report it by income band and age if you can. If it tracks income, so does everything downstream.

4. Verified active rate

Of participants with a connected source, the share recording activity above a threshold on a threshold of days: for example, 5,000 steps on 15 or more days in the month, which is the shape of UnitedHealthcare’s daily goal [12]. This is the first KPI in the stack that describes behaviour rather than enrolment.

The best evidence on what incentives do to it is RAND Europe’s analysis of 422,643 Vitality members in the US, UK and South Africa from 2015 to 2018. Taking up the Active Rewards with Apple Watch benefit, where the member gets the device up front and repays it if activity lapses, was associated with a 34% increase in tracked activity days, about 4.8 more days a month [8]. Two caveats travel with that number: it is an association among people who chose the benefit, and the outcome is tracked days, not health. It is still the largest verified-behaviour dataset in the field, and a 34% lift in verified active days is a reasonable upper bound for what a well-designed, loss-framed incentive can do to this KPI.

5. Sustained engagement

The share of a starting cohort still verified-active in month three and month six. Cohort-based, never cumulative: cumulative registrations only go up.

This is where the word “engagement” needs disambiguation, because three industries use it for three things. Gallup’s employee engagement, 21% globally and 31% in the US and Canada in 2025, measures attitude to work [13]. Vendor engagement, such as Personify Health’s reported 51% average across some 20 million lives, measures activity in the vendor’s app by the vendor’s definition [9]. Clinical engagement means a person is actively managing a condition. A wellness scorecard should say which it means, and a vendor figure without a denominator, a window and a definition of “active” cannot be compared with anything.

The realistic shape of the curve is known from consumer health apps, where the 90-day cliff is the norm. A program that holds half its month-one verified actives at month six is doing well; one that reports 80% engagement without saying over what period is reporting something else.

6. Selection index

The difference between participants and non-participants on baseline risk, measured before the program starts: prior-year claims, screening biometrics, or self-reported health. Report it as a ratio or a gap. It is the single most important number on the scorecard and the one most often missing.

The Illinois study is the reason. In the year before the program, participants had lower medical spending and healthier behaviours than non-participants [2]. Any comparison of participants with non-participants would therefore have shown the program “working.” The randomised comparison showed no effect on spending at 12 or 24 months, no effect on absenteeism, and confidence intervals tight enough to rule out 84% of previously published savings estimates [2]. What the program did move: screening participation, the share with a primary care physician (92.2% against 86.1% at 24 months), and health beliefs by a small margin [14].

If a scorecard cannot report a selection index, its outcome numbers are uninterpretable, and the funder should treat them as such. If it can, and the index is large, the program’s first job is the connected-device rate, because that is where the selection is entering.

7. Outcome attainment rate

The share of participants reaching a defined target, measured with a validated device or laboratory result within a validity window. Borrow the construction from CMS’s ACCESS model, which pays for exactly this: at least 50% of aligned beneficiaries must meet every required target for full payment, targets are defined as either control (systolic blood pressure below 130 mmHg, HbA1c below 6.5% for prediabetes) or minimum improvement (a 15-point reduction in systolic pressure, 5% weight loss), blood pressure and weight readings expire after 15 days, and self-reported values are not accepted except for weight and questionnaires [10]. We wrote up the whole model in three governments just made wearable data a regulated input.

Adopting a payer’s definition does two things for an employer program. It makes the metric legible to the health plan, which increasingly runs the same measure. And it forces the discipline that the trials found missing: BJ’s participants reported exercising more while their measured cholesterol, blood pressure and BMI did not change [3]. Self-report improved. Attainment did not.

8. Cost per verified engaged member

Total incentive and program spend divided by the number of sustained verified actives. The 2025 median incentive was $600 per employee, and three-quarters of employers planned to hold or raise it [5]. Divide that by a participation rate and you get a cost per participant. Divide it by the sustained verified active count and you get the number a finance team can actually use, because it prices the behaviour rather than the sign-up.

What the economics should not include is claims savings, at least not within the horizon of a normal contract. Both randomised trials found none at 18 months, two years or three years [2][3][4]. A program that is justified on claims savings inside that window is being justified on selection. The defensible cases are the ones Discovery and John Hancock actually make, retention and engagement, and we have examined them against the auto-insurance record in health insurance is copying auto telematics.

What good looks like

The last row is deliberate. A program that claims claims savings inside three years is a program that has not read the trials.

The one-page scorecard

For a CFO or a plan sponsor, the report is eight lines and two footnotes.

1. Eligible population, and activation rate.
2. Participation rate, with the incentive type and amount printed beside it.
3. Connected-device rate, split by wearable and phone, and by income band where available.
4. Verified active rate at the program’s stated threshold.
5. Sustained engagement for the oldest cohort, month 6 over month 1.
6. Selection index, with the baseline measure named.
7. Outcome attainment rate, with the target, the measurement method and the validity window.
8. Cost per verified engaged member.

Footnote one: no claims-savings estimate is offered, because randomised evidence does not support one within the contract horizon. Footnote two: the definitions above do not change between quarters.

Where we sit

Sahha supplies verified activity and sleep data to programs like these, from phones as well as wearables, so we have a commercial interest in the connected-device rate being measured and in the phone counting toward it. We think the interest and the evidence point the same way. The trials say selection is the problem; selection enters at verification; verification through a device that 54% of adults do not own is selection by design. Counting the phone is the one lever in this list that addresses the finding of both trials rather than restating it.

The short version

Participation measures the incentive: RAND’s medians are 20% with none, 40% with rewards and 73% with penalties. The randomised trials found participants were healthier before they joined and that clinical measures and spending did not change at 18 months, two years or three years. The scorecard that survives that evidence has seven numbers beside participation: activation rate, connected-device rate, verified active rate, cohort-based sustained engagement, a selection index, a device-verified outcome attainment rate on a payer’s definition, and cost per verified engaged member, with no claims-savings line. The ceiling on verified metrics is device ownership at 46%, unless the phone counts, in which case it is 91%.