Workplace Wellness Programs Were Tested on 38,000 Employees. Tenure Did Not Move.

TLDR: Two randomised trials, at BJ’s Wholesale Club and the University of Illinois, tested workplace wellness programs against tenure, absenteeism and job performance across roughly 38,000 employees, found nothing, and the interventions that did move turnover all altered the job itself.

BJ’s Wholesale Club measured 80 outcomes. Two of them moved.

Those eighty outcomes belong to a trial at BJ’s Wholesale Club covering 32,974 employees. Twenty worksites carrying 4,037 employees received an eight-module programme on nutrition, physical activity and stress reduction delivered by registered dietitians, while 140 control worksites carrying 28,937 employees received nothing. Randomisation ran at the worksite level, the design that captures the culture effect employers say they are buying.

That is what a workplace wellness program usually contains: health risk questionnaires, biometric screening, and coached modules on nutrition, activity and stress, bought as a bundle and offered to everyone. The design question is whether the bundle changes anything an employer is paying it to change.

The programmes themselves are near-universal. In the 2025 employer survey run by KFF across 1,862 employers with ten or more workers, 83 per cent of larger firms offered health and wellness promotion, 53 per cent health risk assessments and 43 per cent biometric screening. Retention is much of why they are bought. In June 2026 the Bureau of Labor Statistics counted 3.2 million American quits anyway.

The results appeared in the Journal of the American Medical Association in April 2019. After eighteen months, treated employees reported regular exercise at 69.8 versus 61.9 per cent, an adjusted difference of 8.3 percentage points, and active weight management at 69.2 versus 54.7 per cent, a difference of 13.6 points. That is the whole of the positive finding. The other twenty-seven self-reported outcomes stayed flat, as did all ten clinical measures, including blood pressure, cholesterol and body mass index, and all thirty-eight measures of health care spending and use.

The fourth domain held three outcomes, absenteeism, tenure and job performance. All three were flat. Song and Baicker published a three-year follow-up in Health Affairs in 2021, with twenty-five treatment worksites and 135 controls. The behaviour gains held. On absenteeism, tenure or job performance, three years produced no detectable improvement either.

Illinois randomised nearly 5,000 people into workplace wellness programs and got the same answer

A second team ran the experiment differently and landed in the same place. At the University of Illinois at Urbana-Champaign, Jones, Molitor and Reif randomised 4,834 employees individually, 3,300 into eligibility and 1,534 into a control group, with the incentive randomised on top. The screening reward was randomised at zero, one hundred or two hundred dollars, and the total available ran from fifty to three hundred and fifty dollars. Fifty-six per cent of eligible employees took part.

The programme did one thing reliably. It raised health screening rates and kept them raised. Past that, the trial reports no significant causal effect on total medical expenditures, other health behaviours, employee productivity or self-reported health status after more than two years. The authors then did something more useful than another null result. Their confidence intervals rule out 84 per cent of previously published estimates on medical spending and absenteeism, which puts a boundary on the prior literature.

That sentence belongs in a budget meeting. It leaves the earlier studies their integrity while saying that if their reported effects were real, this trial would have seen them.

A gap of $116 a month explains why the observational studies glow

The Illinois team also measured the thing that makes wellness reporting so flattering. Employees who completed the screening and health risk assessment had spent $116.10 a month less on health care in the thirteen months before the study than employees who did not. Average monthly spending in that population was $479, so the people who signed up arrived roughly a quarter cheaper than the people who did not, before the programme had done anything at all.

Every design that compares participants with non-participants, or a company before and after launch, inherits that gap and measures who volunteers. Healthier people, with more slack in the week, walk through the door first, and they were always going to cost less and stay longer. The programme takes credit for a sorting effect it did not create.

This generalises well past wellness. Any voluntary internal offer produces the same artefact, whether it is the mentoring scheme, the leadership academy, the graduate rotation or the tuition benefit. Participants outperform non-participants in the dashboard almost every time, and the effect survives most controls an analyst will think to apply, because the sorting variable is motivation and nobody measures it. Ask who chose in before reading any internal programme report.

Katherine Baicker published $3.27 back per dollar, then ran the experiment

The industry’s headline return figure came from one of the same economists. In 2010, Baicker, Cutler and Song published a Health Affairs meta-analysis pooling thirty-six programmes from thirty-two peer-reviewed publications. It found that medical costs fall by about $3.27 for every dollar spent, and absenteeism costs by about $2.73. Those two figures have been quoted in vendor material ever since.

The paper flagged its own weaknesses. It pooled three tiers of design, from randomised trials down to studies with post-intervention data and no baseline, and it warned that early-adopting firms had the highest expected returns, that almost every study came from a large employer, and that publication bias favours high-return findings. Baicker then co-designed both of the randomised trials that came back null.

That sequence is the strongest thing in this evidence base. An author with a professional stake in the 2010 result built the experiments that constrained it and let the intervals sit where they fell. Employers can hold both papers at once, since the 2010 number describes the firms that chose to be studied while the trials describe what happens once the choosing stops.

Three interventions did move turnover, and each one changed the work

The null results stop short of everything carrying a wellbeing label. Every intervention below reached into the conditions of the job itself.

The Work, Family and Health Network ran Support Transform Achieve Results, known as STAR, as a randomised field experiment on turnover at a Fortune 500 firm, assigning fifty-six work groups in an IT division to either supervisor-support training with team-level schedule redesign or to usual practice. Erin Kelly, who co-led the study, puts employees in STAR groups at 40 per cent less likely to leave voluntarily over nearly three years, a magnitude the published abstract gives only as a direction, with burnout and psychological distress moving alongside. Hours worked stayed the same. Control over the hours changed.

Manager capability shows the same pattern at a far smaller dose. A cluster-randomised trial published in The Lancet Psychiatry in 2017 gave 128 Australian fire and rescue managers a four-hour mental health course called RESPECT. Six months later, work-related sickness absence had fallen 0.28 percentage points in the trained group and risen by the same amount in controls, worth 6.45 hours per employee per six months, and the authors calculated a return of £9.98 for each pound spent.

Schedule predictability was tested at Gap, Inc. across twenty-eight stores and 2,331 employees between September 2015 and August 2016, nineteen treated and nine held as controls. Median sales rose 7 per cent and labour productivity 5 per cent. The retention result was mixed, and the report says so. Overall retention held steady, while the average tenure of associates who quit in treatment stores fell by 10.8 months, which the authors read as experienced staff staying longer. The study is more useful for being clear about that difference.

Set against that, the individual-level material keeps failing. William Fleming compared participants and non-participants in 2024 across 46,336 workers in 233 organisations in the United Kingdom, covering resilience training, mindfulness and wellbeing apps. On multiple subjective wellbeing indicators, participants appeared no better off, and Fleming reads the result through job demands and resources, where the interventions supply nothing that answers the demand.

Resignation letter searches outnumber the whole wellness vocabulary fourteen to one

Edward Galle pulled twelve months of United States search volume through the Google Ads keyword planner, covering August 2025 to July 2026, and used the median month per term. One term, “work life balance”, ran to roughly thirty-seven times its own baseline in a single month, which would have pulled an average out of shape. Five wellness and wellbeing terms together reach a median 5,235 searches a month. The single query “resignation letter” reaches 74,000, roughly fourteen times the whole wellness vocabulary combined.

Search volume measures curiosity rather than intent, and it should be read that way. Most of that 74,000 is people who want a template. The vocabulary people reach for when work goes wrong is still the vocabulary of pay, exit and hours. The Search Console property behind this site gives a smaller version of the same picture. Across all 179 queries it surfaced for in the twelve months to 24 August 2026, none carried a wellness, wellbeing or benefits term, while pay benchmarking, permit status and team stress all appeared.

The 33 per cent replacement cost is a vendor number, and the auditable one is 21.4

Almost every business case for retention spending rests on a cost-of-turnover figure. The claim that replacing an employee costs 33 per cent of annual salary traces to the Work Institute’s 2017 Retention Report, quoted at $15,000 for a worker on a $45,000 median salary. No published derivation of the 33 per cent accompanies it.

One figure does come with a stated method. Heather Boushey and Sarah Jane Glynn reviewed thirty case studies across eleven research papers for the Center for American Progress in 2012 and found a median cost of 21.4 per cent of annual salary, falling to 20.7 per cent once physicians and executives are excluded and 16.1 per cent for jobs paying under $30,000. Senior and very highly paid roles ran as high as 213 per cent, which skews every average upward.

Read the vintage before quoting it. Those eleven papers appeared between 1992 and 2007, so an operator citing 21.4 per cent in 2026 is citing evidence between nineteen and thirty-four years old. It remains the best-documented figure available and should travel with its age attached. For the seats this firm recruits into, the percentage framing breaks down anyway. When a qualified person for pharmacovigilance or a person responsible for regulatory compliance leaves, the cost is the period a named regulatory control sits weakly staffed, and that gets priced in inspection findings.

The Americans with Disabilities Act incentive limit has been blank since 2019

A second reason to look hard at a wellness renewal sits in the regulations. Employers running screenings or health risk assessments in the United States are asking disability-related questions and conducting medical examinations, which the Americans with Disabilities Act permits only where participation is voluntary. In 2016 the Equal Employment Opportunity Commission set the boundary of “voluntary” at a 30 per cent incentive. AARP sued. The District Court for the District of Columbia vacated the incentive provisions on 20 December 2017, holding that the agency had not justified the limit it chose.

The commission then removed the incentive section effective 1 January 2019, together with the parallel provision under the Genetic Information Nondiscrimination Act. Both now read as reserved in the Code of Federal Regulations, where section 1630.14(d)(3) is reserved and so is section 1635.8(b)(2)(iii). Replacement rules proposed in January 2021, which would have capped incentives at a de minimis level such as a t-shirt, were withdrawn under a regulatory freeze before publication. In July 2024 the commission returned to strip the matching appendix discussion, effective 5 July 2024, and offered no signal of a replacement.

Meanwhile the health-plan side of the same programme is governed by a rule that still carries a number. The 2013 final regulations on nondiscriminatory wellness programmes, effective 2 August 2013, cap the reward for a health-contingent programme at 30 per cent of the cost of coverage, rising to 50 per cent for tobacco cessation. So a benefits team can build an incentive squarely compliant under one federal regime while the other has offered no standard for more than seven years, an unresolved exposure attached to a programme the trials say does not retain anyone.

Employers outside the United States face the same point in a different shape. The International Organization for Standardization published ISO 45003 on 8 June 2021, setting out guidelines for managing psychosocial risk inside an occupational health and safety management system, and it points at work organisation rather than individual coping. Switzerland has priced the gap through the 2022 Job Stress Index, which is built on a representative survey of 3,022 working people and puts the annual cost of work-related stress to the Swiss economy at around 6.5 billion francs.

Audit the centre before the perimeter renewal lands

Four things are worth doing this week, none of them requiring a new budget line.

Pull twenty-four months of leaver data and cut it by direct manager rather than by department. STAR and the RESPECT training, the two interventions with the cleanest effects, both operated at supervisor level, so that is where a signal will surface. Then convert the wellness renewal quotation into training days. The Lancet Psychiatry trial moved sickness absence on four hours per manager, which makes that comparison quick.

Next, put a rule on turnover numbers. Any cost-of-turnover figure that reaches a slide names its document, its method and its year, or gets replaced with 21.4 per cent and its 1992 to 2007 vintage beside it. Finally, ask the broker to state the disability-law and genetic-information position in writing for any incentive above zero. A broker who answers 30 per cent is quoting a rule removed in 2019.

The pattern holding this together is unglamorous. Programmes bolted onto the outside of a job leave the job unchanged, and employees decide about the job. Edward Galle has argued before that non-financial benefits carry real weight in retention, and the trial evidence sharpens that considerably. The benefits that hold people are the ones that redesign hours, scope and supervision. For regulated employers weighing what keeps a scarce compliance specialist in a seat, see how we work with regulated employers before the next renewal cycle.

References

1. Song Z, Baicker K. Effect of a Workplace Wellness Program on Employee Health and Economic Outcomes: A Randomized Clinical Trial. JAMA, 2019;321(15):1491-1501. https://jamanetwork.com/journals/jama/fullarticle/2730614
2. Song Z, Baicker K. Health And Economic Outcomes Up To Three Years After A Workplace Wellness Program: A Randomized Controlled Trial. Health Affairs, 2021. https://www.healthaffairs.org/doi/10.1377/hlthaff.2020.01808
3. Jones D, Molitor D, Reif J. What do Workplace Wellness Programs do? Evidence from the Illinois Workplace Wellness Study. Quarterly Journal of Economics, 2019;134(4):1747-1791. https://pubmed.ncbi.nlm.nih.gov/31564754/
4. Jones D, Molitor D, Reif J. What do Workplace Wellness Programs do? Evidence from the Illinois Workplace Wellness Study. NBER Working Paper 24229. https://www.nber.org/system/files/working_papers/w24229/w24229.pdf
5. Baicker K, Cutler D, Song Z. Workplace Wellness Programs Can Generate Savings. Health Affairs, 2010;29(2):304-311. http://publichealth.lacounty.gov/diabetes/docs/National_DPP_Research_Articles/Baicker_Workplace_Wellness_Programs_Can_Generate_Savings_2010.pdf
6. Moen P, Kelly EL, Lee SR, et al. Can a Flexibility/Support Initiative Reduce Turnover Intentions and Exits? Results from the Work, Family, and Health Network. Social Problems, 2017;64(1):53-85. https://academic.oup.com/socpro/article-abstract/64/1/53/2753363
7. Moen P, Kelly EL, Fan W, et al. Does a Flexibility/Support Organizational Initiative Improve High-Tech Employees’ Well-Being? Evidence from the Work, Family, and Health Network. American Sociological Review, 2016;81(1). https://www.asanet.org/wp-content/uploads/pr_asr_feb_2015_moen_kelly_study.pdf
8. MIT Sloan School of Management. When good jobs go bad. https://mitsloan.mit.edu/ideas-made-to-matter/when-good-jobs-go-bad
9. Milligan-Saville JS, Tan L, Gayed A, et al. Workplace mental health training for managers and its effect on sick leave in employees: a cluster randomised controlled trial. The Lancet Psychiatry, 2017. https://www.thelancet.com/journals/lanpsy/article/PIIS2215-0366(17)30372-3/fulltext
10. Williams JC, Lambert SJ, Kesavan S, et al. Stable Scheduling Increases Productivity and Sales: The Stable Scheduling Study. https://worklifelaw.org/publications/Stable-Scheduling-Study-Report.pdf
11. Fleming WJ. Employee well-being outcomes from individual-level mental health interventions: Cross-sectional evidence from the United Kingdom. Industrial Relations Journal, 2024;55(2). https://onlinelibrary.wiley.com/doi/10.1111/irj.12418
12. Boushey H, Glynn SJ. There Are Significant Business Costs to Replacing Employees. Center for American Progress, 16 November 2012. https://www.americanprogress.org/wp-content/uploads/sites/2/2012/11/CostofTurnover.pdf
13. HR Dive. Study: Turnover costs employers $15,000 per worker (reporting the Work Institute 2017 Retention Report). https://www.hrdive.com/news/study-turnover-costs-employers-15000-per-worker/449142/
14. Equal Employment Opportunity Commission. Removal of Final ADA Wellness Rule Vacated by Court. Federal Register, 20 December 2018. https://www.federalregister.gov/documents/2018/12/20/2018-27539/removal-of-final-ada-wellness-rule-vacated-by-court
15. Equal Employment Opportunity Commission. Removal of ADA Appendix Sections Related to Removal of Final ADA Wellness Rule Vacated by Court. Federal Register, 5 July 2024. https://www.federalregister.gov/documents/2024/07/05/2024-14606/removal-of-ada-appendix-sections-related-to-removal-of-final-ada-wellness-rule-vacated-by-court
16. Electronic Code of Federal Regulations. 29 CFR 1630.14, Other acceptable examinations and inquiries. https://www.ecfr.gov/current/title-29/subtitle-B/chapter-XIV/part-1630/section-1630.14
17. Electronic Code of Federal Regulations. 29 CFR 1635.8, Acquisition of genetic information. https://www.ecfr.gov/current/title-29/subtitle-B/chapter-XIV/part-1635/section-1635.8
18. Departments of Labor, Health and Human Services, and the Treasury. Incentives for Nondiscriminatory Wellness Programs in Group Health Plans. Federal Register, 3 June 2013. https://www.federalregister.gov/documents/2013/06/03/2013-12916/incentives-for-nondiscriminatory-wellness-programs-in-group-health-plans
19. Baird Holm LLP. EEOC Withdraws Proposed Wellness Plan Incentive Rules. https://www.bairdholm.com/blog/eeoc-withdraws-proposed-wellness-plan-incentive-rules/
20. International Organization for Standardization. ISO 45003:2021, Occupational health and safety management, Psychological health and safety at work, Guidelines for managing psychosocial risks. https://www.iso.org/standard/64283.html
21. Gesundheitsförderung Schweiz / Die Mobiliar. Job Stress Index 2022. https://www.mobiliar.ch/studie/job-stress-index
22. KFF. 2025 Employer Health Benefits Survey, Summary of Findings. https://files.kff.org/attachment/Employer-Health-Benefits-Survey-2025-Annual-Survey-Summary-of-Findings.pdf
23. US Bureau of Labor Statistics. Job Openings and Labor Turnover Survey, June 2026. Released 4 August 2026. https://www.bls.gov/news.release/jolts.nr0.htm
24. Google Ads API, keyword planner historical metrics, United States, English, Google Search network, twelve months to July 2026. Pulled 25 August 2026. Edward Galle first-party data.