Does exercise reduce health insurance claims?

The short answer

Physical activity reduces disease risk. That is not seriously disputed. Whether a programmethat encourages physical activity reduces an insurer's claims is a different question with a much weaker answer, and the gap between those two statements is where most of the wellness industry's marketing lives.

The randomised evidence on workplace wellness found no effect on medical spending. The insurer data that does show lower admissions shows it only among highly engaged members, and only after two to three years. What an insurer is buying is not a benefit to offer; it is sustained, verified activity among members who were previously inactive. Very little of the available evidence demonstrates that an app alone produces that.

What the randomised trials found

Two large randomised trials of workplace wellness programmes report the same uncomfortable result.

The Illinois Workplace Wellness Study found no significant causal effect on total medical spending, biometric markers, diagnoses or utilisation. Its confidence intervals were tight enough to rule out 84% of previously published estimates of savings. It also found that participants had lower spending and healthier behaviours before the programme began — which is the single most important sentence in this literature, because it explains why so many non-randomised studies report savings. [1]

The BJ's Wholesale trial, randomised across 160 worksites over 18 months, raised self-reported regular exercise by 8.3 percentage points and active weight management by 13.6. It changed nothing measurable in BMI, blood pressure, cholesterol, absenteeism, healthcare use or spending. [2]

Self-reported exercise rose while clinical markers did not. Anyone designing a programme on self-logged activity — AdimFit included — has to sit with that finding rather than around it.

What exercise does to the risk factors

The physiological evidence is much stronger than the programme evidence, and it is worth separating the two rather than letting the first borrow credibility for the second.

Across 93 randomised trials, endurance training lowered blood pressure by 3.5/2.5 mmHg overall and by 8.3/5.2 mmHg in hypertensive adults. Isometric resistance training produced 10.9/6.2 mmHg. [3]

On steps, benefit accrues fastest at the bottom of the range and levels off around 7,000 a day. Against a 2,000-step baseline, 7,000 steps was associated with 47% lower all-cause mortality, 25% lower cardiovascular disease, 14% lower type 2 diabetes, 22% lower depression and 28% fewer falls — with the authors flagging low certainty for several of those outcomes. [4] An earlier meta-analysis of 15 cohorts put the mortality plateau at 6,000–8,000 steps for adults over 60 and 8,000–10,000 for those under 60. [5]

What inactivity costs

Roughly one in eight dollars of US healthcare spending is associated with inadequate physical activity: 11.1% (95% CI 7.3–14.9) on the 2015 estimate, [6] and 12.6% (10.5–14.3) — about $205 billion — on the 2025 update, which puts the excess at $2,025 a year for inactive adults and $1,355 for the insufficiently active, adjusted for smoking, obesity, age and insurance status. [7]

These are attributions from national survey data, not causal estimates. They describe the size of the prize. They do not establish that any particular intervention can claim a share of it.

What insurer data shows

Across 948,974 members of a South African health insurer, highly engaged members — 9.5% of the book — had lower cost per admission, shorter stays and fewer admissions: 7.4% lower for cardiovascular disease, 13.2% for cancers, 20.7% for endocrine and metabolic disease. Members with low or no engagement showed no cost benefit at all. [8]

A five-year follow-up of 304,054 members used electronically documented gym visits rather than self-report. Members who stayed highly active through years one to three had fewer admissions in years four and five (20.7% against 22.2%), and two extra gym visits a week corresponded to 13% lower odds of admission (OR 0.87, 95% CI 0.80–0.95). [9]

Neither study is randomised, so neither excludes the selection effect Illinois measured directly. What the second one does establish is timing: claims responded in years four and five, not year one.

Two conditions recur in every dataset where savings appear: the activity was verified rather than self-reported, and the measurement window was long enough for claims to move. Any ROI figure offered on a shorter horizon than that is describing something else.

Can an app move activity at all?

A systematic review and meta-analysis of 28 randomised trials (n = 7,454) of smartphone apps and activity trackers, measured against true and active controls, found a pooled effect of SMD 0.350 (95% CI 0.236–0.465). Among the 21 trials that counted steps directly, the raw pooled difference was +753 steps a day (95% CI 440–971). [11]

Personalisation and message-driven design were significantly more effective in both subgroup analysis and metaregression. That is the category AdimFit sits in, which is a reason to expect the upper end of the distribution rather than a reason to claim a different number.

It is worth being concrete about what +753 buys. A member at 3,500 steps a day moves to roughly 4,250 — real improvement on the steepest part of the dose-response curve, where marginal steps are worth the most, and short of the 6,000–8,000 plateau. An app is a partial dose. Anything claiming otherwise is selling.

For comparison, a UK programme rewarding verified activity raised annual active days by 56% overall and by 554% among previously low-active members, sustained over two years — though it measured activity rather than claims, and it was observational. [10]

Correction

What this evidence does not support

Stated plainly, because these are the claims the category makes and we are not going to make them:

• That participants in a wellness programme cost less because of the programme. Healthy-user selection explains this, and Illinois measured it directly.
• Any first-year claims-savings return. Claims responded with a two-to-three-year lag in the insurer data, and not at all at 18 months in the randomised trial.
• That apps add 1,850 steps a day. That figure is a unit conversion of a standardised effect. The directly pooled estimate is +753.
• That vendor-published return-on-investment figures represent what any particular programme will achieve. They are evidence the category can work, with their caveats attached.

How this page was made

Every figure is traceable to a numbered source below. Where a study is observational it is labelled as such. Where a figure comes from a vendor rather than a peer-reviewed journal, it is not used. Where an earlier version of this page was wrong, the correction stays visible above rather than being edited away.

This page is drawn from a longer internal review that also contains commercial analysis. That part is not published. The evidence is public; what we intend to do about it is not.

Common questions

Sources

1. 01

   Jones D, Molitor D, Reif J (2019). What do workplace wellness programs do? Evidence from the Illinois Workplace Wellness Study.

   The Quarterly Journal of Economics 134(4): 1747–1791 · doi:10.1093/qje/qjz023

   Randomised. The strongest evidence on this page, and the least convenient.

2. 02

   Song Z, Baicker K (2019). Effect of a workplace wellness program on employee health and economic outcomes: a randomized clinical trial.

   JAMA 321(15): 1491–1501 · PMID 30990549

   Cluster-randomised across 160 worksites. 18-month follow-up.

3. 03

   Cornelissen VA, Smart NA (2013). Exercise training for blood pressure: a systematic review and meta-analysis.

   Journal of the American Heart Association 2(1): e004473 · PMID 23525435

   Meta-analysis of 93 randomised trials, n = 5,223.

4. 04

   Ding D, Nguyen B, Nau T, et al. (2025). Daily steps and health outcomes in adults: a systematic review and dose-response meta-analysis.

   The Lancet Public Health · PMID 40713949

   57 studies, 160,000+ adults. The authors flag low certainty for several outcomes.

5. 05

   Paluch AE, Bajpai S, Bassett DR, et al. (2022). Daily steps and all-cause mortality: a meta-analysis of 15 international cohorts.

   The Lancet Public Health 7(3): e219–e228 · PMID 35247352

   Observational cohorts, ~50,000 people. Establishes the plateau.

6. 06

   Carlson SA, Fulton JE, Pratt M, Yang Z, Adams EK (2015). Inadequate physical activity and health care expenditures in the United States.

   Progress in Cardiovascular Diseases 57(4): 315–323 · PMID 25559060

   National survey attribution. Association, not causation.

7. 07

   Matjasko JL, Carlson SA, Whitfield GP, et al. (2025). Health care expenditures attributable to inadequate physical activity among US adults.

   American Journal of Health Promotion · doi:10.1177/08901171251357128

   Updates Carlson et al. Adjusted for smoking, obesity, age and insurance status.

8. 08

   Patel D, Lambert EV, da Silva R, et al. (2010). The association between medical costs and participation in the Vitality health promotion program among 948,974 members of a South African health insurance company.

   American Journal of Health Promotion 24(3): 199–204 · PMID 20073387

   Cross-sectional, one year. Selection effects not excluded.

9. 09

   Patel DN, Lambert EV, da Silva R, et al. (2011). Participation in fitness-related activities of an incentive-based health promotion program and hospital costs: a retrospective longitudinal study.

   American Journal of Health Promotion 25(5): 341–348 · PMID 21534837

   Five years, electronically documented gym visits. Longitudinal but not randomised.

10. 10

    Hajat C, Hasan A, Subel S, Noach A (2019). The impact of short-term incentives on physical activity in a UK behavioural incentives programme.

    npj Digital Medicine 2: 91 · doi:10.1038/s41746-019-0164-3

    Observational, n = 11,881. Measures activity, not claims.

11. 11

    Laranjo L, Ding D, Heleno B, Kocaballi AB, Quiroz JC, Tong HL, et al. (2021). Do smartphone applications and activity trackers increase physical activity in adults? Systematic review, meta-analysis and metaregression.

    British Journal of Sports Medicine 55(8): 422–432 · PMID 33355160

    28 randomised trials, n = 7,454. The source of the correction logged on this page.