Our Recommendation
Shamiri has a structural advantage over almost every organization in our directory, and it is worth naming plainly: the outcome Shamiri measures is an ultimate outcome. Most organizations we evaluate deliver something (a well, a loan, a bag of food, a job placement) and then hope it produces health or dignity or wellbeing downstream, which is exactly where the measurement usually stops.
Shamiri delivers four group sessions and then measures whether a young person still meets the clinical threshold for depression. There is no downstream link to assume. What makes the score defensible rather than lucky is the depth behind that measurement: 12-plus randomized controlled trials, 18,000-plus participants, 35-plus peer-reviewed publications. Alongside that sit intermediary counterfactuals on academic performance and, more unusually, two dissemination trials totaling over 5,000 participants designed to test whether partners rather than Shamiri could deliver the model and get comparable results.
Shamiri Institute's Fierce Certification score is 100/100 points based on our criteria:
✔ Has Ultimate Outcome Goals (20 pts)
✔ Measures Intermediate Outcomes (15 pts)
✔ Measures Ultimate Outcomes (20 pts)
✔ Measures Intermediate and/or Ultimate Counterfactual (20 pts)
✔ Shows Continual Learning & Adaptation (25 pts)
The Social Problem
An estimated 250 million young people worldwide are struggling with mental health, and mental health is now the leading cause of disability for ages 10 to 24. In low-resource settings, more than 80 percent of young people who need support will never receive any.
The reason is not that treatment does not exist. It is that the delivery model does not. Across most of Africa and the Global South, fewer than one mental health professional serves every 100,000 people. Private therapy in Kenya costs roughly $400 per person, which is beyond household reach for the families Shamiri serves. And the services that do exist were designed around adult clinical models in adult settings that adolescents do not walk into.
The consequences compound over time rather than resolving. According to Shamiri, sixty percent of youth mental health conditions persist into adulthood, affecting education, employment, relationships, and health for decades. Before that, they show up in the years being lived: grades falling, relationships narrowing, and a young person's sense of control over their own life eroding at precisely the age when it should be forming.
Shamiri's framing of this is the sharpest problem statement in our directory: most mental health systems were designed for a world with enough professionals, enough money, and enough infrastructure, and that world does not exist for the majority of young people. The organization was built for the world as it is.
The Solution
Expand who can deliver care, and equip them with something that works.
The tiered provider model (Model). Young adult lay professionals deliver brief interventions. Semi-professionals recruit, train, and support them. Experts handle complex cases. The organization describes this as the right level of care for every level of need, and it is the mechanism that breaks the dependence on a professional workforce that does not exist.
The intervention. Four brief group sessions, delivered in schools where young people already are. The 2020 pilot compared it against a study skills control, which is a meaningful choice: the comparison was not nothing, but a plausible alternative use of the same school hours.
The technology layer (Tech). shamiriOS manages operations, quality, and reporting. shamiriAI supports training and supervision, with an automated fidelity monitoring tool now pilot-validated. The stated purpose is transfer of ownership: partners and governments delivering Shamiri-quality care through their own systems and their own people.
Two scaling routes. Shamiri Hubs, which the organization runs directly, each serving up to 20,000 young people annually across Nairobi, Kiambu, Kajiado, Murang'a, Kisumu, Kakamega, and Ethiopia. And Shamiri Partners, where governments and local partners run Hubs independently, powered by Shamiri's technology and infrastructure.
Mapped against a Theory of Change, the chain is unusually short: expand the provider pool, deliver brief structured sessions in schools, symptoms fall. Every activity maps to that chain, and the chain terminates at an ultimate outcome rather than passing through several assumed links. The extension claims about livelihoods, education, and social connection are a second, longer chain that the organization is currently testing rather than asserting as established, though the Impact page does not always make that distinction visible.
Key Outputs
- 285,000-plus youth served since 2021, reported as 250,000-plus on the Impact page, currently serving over 100,000 annually
- 70,000-plus sessions held across 500-plus schools in 10-plus counties
- 2,100-plus providers trained
- $7 per youth, against $25 for peer therapy, $45 for group therapy, $175 for lay cognitive behavioral therapy in low- and middle-income countries, and $400 for private care in Kenya. Stated goal of $3 per client
- 12-plus randomized controlled trials, 18,000-plus trial participants, 35-plus peer-reviewed publications
- Scaling structure: Hubs in six Kenyan counties plus Ethiopia, each serving up to 20,000 annually; 5-plus county governments and 10-plus partner organizations delivering independently
- Stated target: 10 million young people by 2031
- Government partnerships with the Kenyan Ministry of Education, Ministry of Health, Nairobi County, and Murang'a County
Key Intermediate Outcomes
Partner delivery is the most interesting measured intermediary outcome, and it is the one many organizations never test. Dissemination Trial 1.0 in 2022 (N=3,473) and Dissemination Trial 2.0 in 2023 (N=1,576) were designed to answer whether partners could deliver the model with comparable outcomes across contexts. That is a non-inferiority question about the organization's own indispensability, asked with over 5,000 participants, and answered in the direction that makes Shamiri less necessary rather than more.
Academic performance was measured with a control group in the 2020 Behavior Therapy pilot, which randomized 51 Kenyan adolescents against a study skills comparison and found improvement in grades alongside symptom reduction. The Impact page quantifies this as half to a full letter grade.
Cost is measured and benchmarked, at $7 per youth against a published comparison set. This is a benchmark rather than a counterfactual, but it is transparent about its comparators in a way most cost claims are not.
Provider capacity and fidelity. 2,100-plus providers trained, with shamiriAI pilot-validated as an automated fidelity monitoring tool in JMIR AI (2026, N=52). The published abstract is notably candid: it reports that underperformance was concentrated on holistic inferential dimensions, specifically Required Contents and Clarity, while structured detection tasks approach operational utility, and attributes the gap to diagnosable misalignments in rubric interpretation and prompt design. Publishing where your own artificial intelligence tool falls short, in a peer-reviewed venue, is unusual behavior.
Key Ultimate Outcomes
Depression and anxiety are measured, with randomized controls, in peer-reviewed journals.
The 2021 JAMA Psychiatry trial randomized 413 Kenyan adolescents and found the Shamiri intervention produced greater reductions in depressive symptoms (Cohen's d = 0.35) and anxiety symptoms (d = 0.37) at post-treatment, with effects sustained through seven-month follow-up. The 2020 Behavior Therapy pilot randomized 51 adolescents against a study skills control and found d = 0.32 for depression and d = 0.54 for anxiety.
Shamiri separately reports on its impact page: 80 percent of young people with clinically elevated depression and 82 percent with clinically elevated anxiety no longer meet the clinical threshold after four sessions, with effects sustained across the full school year.
A word on magnitude. Effect sizes of 0.32 to 0.54 are small to moderate by conventional standards, and the organization's framing that outcomes are comparable to expert-delivered therapy is accurate in that brief interventions and full therapy tend to land in similar ranges in this literature. The case for Shamiri is not that the effect is unusually large. It is that a comparable effect can be delivered by someone who is not a psychologist, to a population that would otherwise receive nothing. That is the right claim, and the organization makes it.
Three ultimate outcomes are claimed without evidence on the page. Improved sense of control over life outcomes (agency), stronger peer relationships and increased sense of belonging (social connection), and better preparedness for work and life. All three are named as ultimate outcomes in the Fierce Certification framework, all three appear on the Impact page under a heading stating that when mental health improves, everything improves, and none carries a figure, an instrument, or a citation. Given that the organization has published 35 papers and cites effect sizes precisely elsewhere, the contrast is conspicuous. Either the measurements exist and should be shown, or they do not and the claims should be labeled as hypotheses under test.
One published result complicates the picture, and Shamiri published it. A five-arm trial during the COVID period found that Shamiri's effects held steady against its own prior trials, while the study skills control produced 31 percent greater anxiety reduction and 60 percent greater depression reduction than controls in earlier trials. The authors' conclusion is that control conditions teaching contextually relevant life skills may themselves produce mental health benefits. That is an organization reporting that its comparison group caught up with it, and it is the reason we would want to see the next trial before treating the effect size as settled.
Continual Learning & Adaptation
The learning function is structural, not occasional. Shamiri operates two named systems: an Evidence Engine that establishes what works through randomized trials, pre-registered protocols, clinical outcomes, and long-term follow-up; and an Optimization Engine that learns how to scale through A/B testing, non-inferiority trials, mixed methods, and monitoring and evaluation.
The evidence roadmap is a visible progression, not a collection. Fifty-one participants in 2018 to 413 in 2019 to 1,252 in 2021 to 3,473 in 2022 to 1,576 in 2023 to 657 in 2024 to 5,250 in 2025. Crucially, the questions changed at each step rather than the sample size alone: the 2021 multi-arm trial moved from "does it work" to "why it works" by identifying what drives improvement. That is an organization testing its own mechanism rather than re-proving its own effect.
It publishes its own four conditions and tests against them. Good enough (does it work), big enough (can it solve the problem at scale), simple enough (can it be delivered with fidelity), cheap enough (is it affordable at scale). Stating in advance the conditions under which your model would fail to scale, and then running trials against each, is the closest thing in this directory to a published set of stopping criteria.
Current open questions are stated as open. Can governments deliver care effectively. Can supervision be automated without losing fidelity. Do mental health interventions drive broader long-term life outcomes. Can we intervene earlier and effectively. Each is described with the language of early evidence rather than established finding.
It publishes unflattering technical results. The shamiriAI validation paper reports precisely where the tool underperformed and why, in a peer-reviewed journal, at a stage when the organization could simply have waited for better numbers.
Learning is shared as a program. Thirty-five-plus publications across the highest-tier journals in the field, an open publications index, an annual Summit and a convening called The Gathering, an Academy, and a webinar series. The organization describes itself as working from Africa for the world, and the research output supports that framing rather than decorating it.
Further Reading
Organization Sources
- Shamiri Institute homepage, source of the 285,000 youth served, 80 percent recovery, 2,100 providers, and $7 cost figures. shamiri.co
- Shamiri, Science, source of the Evidence Engine and Optimization Engine framework, the four conditions, the evidence roadmap, and the trial and publication counts. shamiri.co/science
- Shamiri, Impact, source of the 80 and 82 percent recovery rates, the cost comparison set, the Hub structure, and the claimed livelihoods, academic, social support, and agency outcomes. shamiri.co/impact
- Shamiri, Model and Tech, source of the tiered provider structure, shamiriOS, and shamiriAI. shamiri.co/model and shamiri.co/tech
- Shamiri, Research index and Impact Dashboard. shamiri.co/research and impact dashboard
Peer-Reviewed Evidence
- Osborn, T., et al. "Randomized controlled trial of the Shamiri intervention." JAMA Psychiatry, 2021, N=413. Source of d = 0.35 for depression, d = 0.37 for anxiety, and 7-month sustainment. Study page
- "Pilot randomized trial of Shamiri in Kibera." Behavior Therapy, 2020, N=51. Source of d = 0.32 for depression, d = 0.54 for anxiety, and academic performance gains against a study skills control. Study page
- "shamiriAI for Supervision Pilot Validation Study." JMIR AI, 2026, N=52. Source of the automated fidelity monitoring validation and its reported shortcomings. Study page
- Shamiri 3.0 multi-arm mechanism trial, 2021, N=1,252; Dissemination Trial 1.0, 2022, N=3,473; Dissemination Trial 2.0, 2023, N=1,576; Livelihoods and long-term impact trial, 2024, N=657. Summarized on the Science page evidence roadmap.
Independent Coverage and Recognition
- Nature, "How students and grandparents could solve the global mental health crisis."
- Harvard Magazine, "A new prescription for youth mental health."
- Elevate Prize Foundation, 2026 winner, Tom Osborn and Shamiri. elevateprize.org
Written by
