Research · October 2026
The research behind Kratos
Our hypothesis in one line: coaching works, but it runs on a coach’s hours, and a studio’s revenue stops at the seats in the room. AI that drafts, with a coach who approves, could lift both. Here is the evidence, the numbers we derived, and where the evidence is thin.
01 · The demand
Nearly 1 in 3 adults worldwide don’t move enough.
The fix is well understood. The bottleneck is delivering it to enough people.
1.8 billion adults, 31%, were insufficiently active in 2022, up about 5 points since 2010. *On the current trend, 35% by 2030.1
Show the numbers
| Year | Share of adults | Note |
|---|---|---|
| 2010 | ~26% | Derived: about 5 points below 2022 |
| 2022 | 31% | About 1.8 billion adults |
| 2030 | 35% | Projection if the trend continues |
The tick marks the earlier year. About 3 in 4 Australian adults2 and 2 in 3 US adults3 don’t do muscle-strengthening activity twice a week, and neither number has moved much.
Show the numbers
| Market | Earlier | Latest | Measure |
|---|---|---|---|
| Australia | 77% (2017–18) | 76% (2022) | Did not meet the muscle-strengthening guideline |
| United States | 68% (2020) | 67% (2024) | 100% minus the share doing muscle-strengthening on 2+ days (31.9%, 32.9%) |
02 · Coaching works
Supervised training came out ahead in 25 of 30 comparisons.
Eleven randomised trials, 621 older adults, the same exercises done with and without an instructor.
25 of the 30 between-group comparisons favoured the supervised group.4
Show the numbers
| Outcome | Effect size (SMD) | Studies | Reading |
|---|---|---|---|
| Muscle strength and power | 0.51 | 5 | Medium, favours supervised |
| Overall balance | 0.40 | 10 | Small to medium |
| Balance test batteries | 0.53 | 4 | Medium |
| Dynamic balance | 0.35 | 5 | Small |
| Proactive balance | 0.24 | 9 | Small |
| Static balance | 0.28 | 6 | Small, not significant |
Two of every three sessions supervised.
The authors’ recommendation for balance and strength programs in older adults. A program counted as supervised when an instructor attended at least two of every three sessions.4
It wasn’t just turning up.
Adherence was similar with and without supervision (80.9% vs 76.8%), so attending more doesn’t explain the gap. The authors point to better technique and higher intensity under supervision.4
A little coaching closes most of the gap
Line up 100 people from each comparison group. How many would the average fully coached person beat on strength and power? 50 means no difference.
vs people who trained entirely alone
89 of 100
vs people who trained mostly alone, with a few coached sessions
55 of 100
Show the numbers
| Comparison | Effect size (SMD) | Trials | As “of 100” (Cohen’s U3) |
|---|---|---|---|
| Fully coached vs entirely alone | 1.24 (95% CI 0.72–1.77) | 2 | 89 |
| Fully coached vs mostly alone with a few coached sessions | 0.12 (95% CI −0.12 to 0.36) | 3 | 55 |
03 · The coach’s ceiling
At the recommended dose, one coach tops out at 21 clients.
Coaching is sold by the hour, and the hours run out. Then the hours pay less than most jobs.
43 hours is the average full-time week for an Australian fitness instructor.5 The arithmetic assumes every hour is coached; programming, check-ins and admin come out of the same hours.
Show the numbers
| Sessions per client per week | Clients per 43 hours |
|---|---|
| 1 | 43 |
| 2 | 21 (1 hour left) |
| 3 | 14 (1 hour left) |
The end of each track is the typical worker’s pay (100%).
Australia compares median hourly earnings5; the US compares median annual wages.6 Each ratio uses the same measure on both sides.
Show the numbers
| Market | Fitness trainers | All workers | Ratio |
|---|---|---|---|
| Australia (hourly, May 2025) | $38 | $47 | 80.9% |
| United States (annual, May 2025) | $47,160 | $50,980 | 92.5% |
04 · The studio’s ceiling
A studio’s revenue is capped by the room before it’s capped by demand.
Revenue is seats × classes × price, and the room fixes the seats. Then members leave before anyone has time to coach them.
From Xponential Fitness’s FY2024 10-K. Club Pilates, the largest Pilates brand by number of studios, is designed for up to 12 people per class.7
Show the numbers
| Brand | Designed capacity per class | Typical size |
|---|---|---|
| Club Pilates | 12 | ~1,500 sq ft |
| Pure Barre | 26 | ~1,500 sq ft |
| BFT | 36 | ~2,500 sq ft |
| YogaSix | 40 | ~2,000 sq ft |
| Rumble (boutique) | ~48 | ~2,500 sq ft |
| CycleBar | 50 | ~2,000 sq ft |
| Rumble (signature) | ~60 | 3,500–4,500 sq ft |
Revenue rose an average of 8%, memberships 6% and facility numbers nearly 4%.8 Much of the growth came with more floor space.
Show the numbers
| Measure | Change in 2024 |
|---|---|
| Revenue | +8% (average) |
| Memberships | +6% |
| Facilities | nearly +4% |
63% of new members abandoned training before their third month; fewer than 4% stayed active for more than 12 continuous months.9
Show the numbers
| Outcome | Share of new members |
|---|---|
| Quit before month three | 63% |
| Quit between month three and twelve | about 33% (derived) |
| Still active after 12 months | fewer than 4% |
05 · Beyond the ceiling
AI can now write a training plan. Clients still need a coach to stick with it.
Software on its own doesn’t keep people. A coach within reach does. And AI is now good enough to do the prep.
An average of 43% of participants dropped out (95% CI 29–57). The ability to contact a health professional was found to increase engagement, and people using apps alongside phone or internet programs were more likely to stay.10
Show the numbers
| Outcome | Share |
|---|---|
| Stayed | 57% |
| Dropped out (pooled) | 43% |
| Dropped out, randomised trials | 40% |
| Dropped out, observational studies | 49% |
Google’s personal health model scored above a sample of experts on both exams, and on real fitness case studies its answers were rated no different from the experts’ own.11
Show the numbers
| Exam | AI model | Experts | Experts |
|---|---|---|---|
| Fitness (99 questions) | 88% | 71% | 5 athletic trainers, 13.8 years’ experience on average |
| Sleep (629 questions) | 79% | 76% | 5 sleep medicine experts, on a 204-question subset |
Top 10 worldwide fitness trends for 2026
- 1Wearable technology
- 2Fitness programs for older adults
- 3Exercise for weight management
- 4Mobile exercise apps
- 5Balance, flow and core strength
- 6Exercise for mental health
- 7Traditional strength training
- 8Data-driven technology
- 9Adult recreation and sport clubs
- 10Functional fitness training
Ranked by about 2,000 exercise professionals. Three of the top ten are about data.12
Our hypothesis
Let AI do the prep: read the data, draft the program change, write the check-in. Let the coach approve every draft and own the relationship. Programming moves off the coach’s clock, and members get coached on the days they’re not in the room.
What a pilot will measure
Coach review time per member per week. New members still training at day 90, against the studio’s own numbers. Check-ins answered within 24 hours. Sessions attended per member per week.
06 · Sources
Sources and method
We use primary sources: government statistics, peer-reviewed papers, company filings and industry bodies. Each source says what we read. Numbers we derived are marked as derived, with the working shown.
World Health Organization (2024). Nearly 1.8 billion adults at risk of disease from not doing enough physical activity. Reporting Strain T et al., The Lancet Global Health. www.who.int/news/item/26-06-2024-nearly-1.8-billion-adults-at-risk-of-disease-from-not-doing-enough-physical-activity
Read: the WHO release in full.
Australian Institute of Health and Welfare. Physical activity (from ABS National Health Survey 2022). www.aihw.gov.au/reports/physical-activity/physical-activity
Read: the web report.
US Office of Disease Prevention and Health Promotion. Healthy People 2030, objective PA-04 (CDC National Health Interview Survey). odphp.health.gov/healthypeople/objectives-and-data/browse-objectives/physical-activity/increase-proportion-adults-who-do-enough-muscle-strengthening-activity-pa-04
Read: the objective page.
Lacroix A, Hortobágyi T, Beurskens R, Granacher U (2017). Effects of supervised vs. unsupervised training programs on balance and muscle strength in older adults: a systematic review and meta-analysis. Sports Medicine 47(11): 2341–2361. doi.org/10.1007/s40279-017-0747-6
Read: the full paper.
Jobs and Skills Australia. Fitness Instructors (ANZSCO 4521), labour market insights. Data from ABS Labour Force Survey 2025, Survey of Employee Earnings and Hours May 2025, Census 2021. www.jobsandskills.gov.au/data/labour-market-insights/occupations/4521-fitness-instructors
Read: the profile page (a legacy ANZSCO page that will not be updated again).
US Bureau of Labor Statistics. Occupational Outlook Handbook: Fitness Trainers and Instructors (May 2025 wage data). www.bls.gov/ooh/personal-care-and-service/fitness-trainers-and-instructors.htm
Read: the handbook page in full.
Xponential Fitness, Inc. Form 10-K for fiscal year 2024, filed with the US Securities and Exchange Commission. www.sec.gov/Archives/edgar/data/1802156/000095017025039541/xpof-20241231.htm
Read: the filing’s business section (about 100,000 of 113,000 characters).
Health & Fitness Association (2025). 2025 Global Fitness Industry Report Shows Record Growth and What’s Next for the Market. www.healthandfitness.org/?p=4214
Read: HFA’s article on its report; not the report’s tables.
Sperandei S, Vieira MC, Reis AC (2016). Adherence to physical activity in an unsupervised setting: explanatory variables for high attrition rates among fitness center members. Journal of Science and Medicine in Sport 19(11): 916–920. doi.org/10.1016/j.jsams.2015.12.522
Read: the abstract only.
Meyerowitz-Katz G, Ravi S, Arnolda L, Feng X, Maberly G, Astell-Burt T (2020). Rates of attrition and dropout in app-based interventions for chronic disease: systematic review and meta-analysis. Journal of Medical Internet Research 22(9): e20283. www.jmir.org/2020/9/e20283/
Read: the abstract and the relevant results.
Cosentino J, Belyaeva A, Liu X, et al. (2025). A personal health large language model for sleep and fitness coaching. Nature Medicine. Exam and expert details from the preprint, arXiv:2406.06474. doi.org/10.1038/s41591-025-03888-0
Read: Google Research’s summary and the arXiv preprint; not the journal version.
American College of Sports Medicine (2025). 2026 ACSM Worldwide Fitness Trends: Future Directions of the Health and Fitness Industry. ACSM’s Health & Fitness Journal. acsm.org/top-fitness-trends-2026
Read: ACSM’s announcement page with the top 10.
Derived numbers
- World 2010 inactivity, ~26%: WHO reports a rise of about 5 points from 2010 to 2022 (31%).
- US adults not strength training: 100% minus the published share who do (31.9% in 2020, 32.9% in 2024).
- Pay ratios: Australia $38 ÷ $47 = 80.9%; US $47,160 ÷ $50,980 = 92.5%.
- Australian part-time share across all jobs, 31%: 100% minus the published 69% full-time share.
- Clients per week: 43 hours ÷ sessions per client, rounded down.
- “Of 100”: Cohen’s U3 from the paper’s effect sizes, Φ(1.24) = 0.89 and Φ(0.12) = 0.55.
- Members quitting between month three and twelve, about 33%: 100% − 63% − just under 4%.
Spotted an error or a better source? Tell us and we’ll correct it.
The plan can be drafted. The relationship can’t.
We’re opening a small pilot with boutique studios and coach teams to test this.