At Youth Impact we’ve spent a decade grappling with a central tension: how do we design programs that not only work in controlled settings but can reach hundreds of thousands of young people through government systems?​ The answer, we’ve found, lies in three interconnected practices: grounding our work in rigorous evidence, maintaining relentless adaptability through systematic testing, and embedding ourselves within—rather than alongside—government structures. This blog discusses what we’ve learned putting these principles into practice.

​

Creating conditions for agency through accurate information
​

We believe that accurate, actionable information is foundational to agency. Young people cannot make informed choices about their health, education, or futures without access to evidence-based knowledge. Yet in many contexts, this information simply doesn’t reach them—or reaches them too late.

​

Our Choices curriculum exemplifies this approach. In Botswana, we work with junior secondary school students to provide clear, factual information about HIV risk in age-disparate relationships. The reality is stark: in settings with high HIV prevalence, relationships between young girls and significantly older partners carry substantially higher infection risk. By providing this information directly—not through scare tactics or moralizing, but through straightforward education about relative risk—we enable young people to make more informed decisions about their relationships and health.

​

Similarly, our ConnectEd tutoring program addresses educational inequalities by ensuring that children falling behind in foundational numeracy receive targeted, phone-based instruction. But it goes further: we train caregivers to support their children’s learning, creating a ripple effect where parents gain the knowledge and confidence to become educational partners. This matters in contexts where caregivers often feel they lack the skills to help their children academically.

​

Information alone doesn’t dismantle structural barriers—economic constraints, social norms, and power imbalances remain. But accurate information removes one critical obstacle to agency: the knowledge gap that prevents young people and their families from navigating the systems and choices available to them.

​

Evidence as the foundation for our programs
​

Everything we do starts with evidence, but in a novel way. We don’t simply review existing research and adopt “best practices.” Instead, we subject every program to rigorous testing in the specific contexts where we work.​When we learned about a successful HIV prevention curriculum in Kenya, we didn’t simply replicate it in Botswana. We adapted the program for the local context and then conducted a randomized controlled trial to verify it worked in this new setting. The results confirmed the impact for certain delivery models, but we didn’t stop there (Angrist et al. 2019).

​

One critical question emerged: who should deliver this sensitive content? Through systematic testing documented in our “messenger matters” research, we found that near-peer instructors—members of government youth service corps just slightly older than the students—were more effective than teachers (Angrist 2019). Young people were more receptive to messages about sexual health and relationships from peers they could relate to than from authority figures in the classroom. This finding fundamentally shaped our implementation model.

​

For ConnectEd, we’ve conducted twelve rounds of A/B testing to optimize the program for cost-effectiveness and scalability (Angrist, Cullen, and Magat 2025). Some tests focused on reducing costs without compromising impact: could bi-weekly 40-minute calls work as well as weekly 20-minute sessions? (Yes, and they reduced scheduling costs.) Could different tutors rotate rather than maintaining the same tutor throughout? (Yes, increasing scheduling flexibility.)

​

Other tests aimed to enhance effectiveness at minimal cost. We discovered that encouraging caregivers to co-lead part of the tutoring call—taking over halfway through, after listening to the tutor model instruction—more than doubled the program’s impact on learning. This modification cost virtually nothing to implement but generated learning gains of 0.25 standard deviations, far exceeding typical educational interventions.​ This systematic experimentation yielded efficiency improvements in 58 percent of tests we conducted—a success rate that compares favorably to the 10-40 percent rate common in the technology sector’s A/B testing culture. The insight here is simple but powerful: the difference between a program that works in one setting and a program that can scale sustainably often comes down to dozens of small optimizations, each tested rigorously.

​

Adaptability as a core competency
​

Evidence isn’t static and we strive not to allow our programming to be either. The iterative A/B testing approach allows us to learn and adapt continuously rather than treating program design as fixed. When we find that rotating tutors works as well as each child having a consistent tutor, we implement that finding, increasing our operational flexibility. When caregiver engagement proves transformative, we build it into our model going forward.

​

This adaptability also applies to how we respond to implementation realities. In the Philippines, we co-designed ConnectEd with the Department of Education from the start, ensuring the program used technology that households already owned (basic mobile phones), aligned with the existing curriculum, fit within manageable teacher workloads, and tapped into established budget lines for teacher airtime allowances. The result: a program now scaling across 20 school division offices because it was designed for government implementation from day one, not retrofitted later.

​

Government partnership as the path to scale
​

This is perhaps our most important operating principle: we only implement through government systems, never parallel to them. In education and health—the sectors where we work—there is simply no path to reaching hundreds of thousands of young people without government buy-in and execution.

​

Too many NGO pilots operate in isolation, demonstrating impact in small-scale settings that can’t translate to government delivery. We take a different approach, embodied in what we call the “three S’s”:

​

• Spark. We pilot programs to demonstrate results and build buy-in. In Puerto Princesa, Philippines, our hands-on support enabled early visible improvements in children’s numeracy. The local superintendent became an enthusiastic champion, and other regions began visiting to understand the model. Quick, tangible results create momentum.

• Support. We provide continuity across political transitions and capacity where government systems are stretched thin. In Botswana, we’ve trained a cohort of master trainers who can support government personnel and seconded regional coordinators into government offices. When Botswana’s Ministry of Basic Education saw multiple different Permanent Secretaries in just a few years, this embedded support maintained program continuity despite leadership turnover.

• Study. We maintain sites where we can experiment and optimize, identifying modifications that reduce costs or enhance effectiveness. Governments benefit from this “development sandbox” without bearing the risk and cost of experimentation themselves.

​

This model requires patience and humility. We’re not the lead boat cutting through water—we’re the stabilizing outriggers enabling government systems to navigate complex implementation challenges. Our role is to make government delivery more effective, not to demonstrate that NGOs can do it better.

​

Scaling without voltage drop
​

One common criticism of development programs is that impact often diminishes at scale—what researchers call “voltage drop” (List 2024). Promising pilots fail to maintain their effectiveness when expanded. Our experience suggests this isn’t inevitable.​ Through continuous optimization, we’ve actually increased effectiveness while reducing costs.

 

Our original Botswana tutoring trial showed learning gains of 0.12 standard deviations. After rounds of A/B testing and refinement, modifications like caregiver engagement now generate gains of 0.20-0.25 standard deviations—more than double the original impact. Simultaneously, operational improvements have reduced per-student costs by 11 percent.​ The key is treating scale-up not as simple expansion but as an ongoing process of learning and adaptation, conducted in genuine partnership with the government systems that will ultimately own and sustain the work.

​

Looking ahead
​

Youth agency develops when young people have both accurate information and systems that respond to their needs. Our work focuses on the first part—ensuring evidence-based knowledge reaches young people at scale—while recognizing it’s only one piece of a larger puzzle.​ What gives me hope is that the model we’re building proves sustainable. In Karnataka, India, 75,000 government teachers across the state are being trained this year to deliver ConnectEd to students in all primary schools, supported by government master trainers and existing education system structures. In Botswana, the government has adopted the tutoring model we tested and optimized. These aren’t NGO programs happening in government schools—they’re government programs, improved through evidence and partnership.

​

Advancing youth agency requires us to work at multiple levels: individual information provision, program optimization, and systems strengthening. Youth Impact’s contribution is demonstrating that rigorous evidence, genuine adaptability, and embedded government partnership can work together—creating conditions where more young people have the information and support they need to exercise agency in their lives.

​

This blog is part of a series by participants at The Agency Summit: Expanding Opportunities to Act, Choose, and Thrive, co-hosted by The Agency Fund and New York University.