Tarjimly, Jacaranda Health, and the AI-Native Model: How Nonprofits Achieve 10x Impact

The term "AI-native" is easy to misunderstand. It does not simply mean using AI tools, any more than a "cloud-native" software company is just one that stores files in Dropbox. An AI-native nonprofit is one where AI is embedded in the organization's fundamental operating model, not layered on top of an existing structure.

The distinction matters enormously in practice. A traditional nonprofit that adopts AI typically uses it to make existing processes faster or cheaper. Staff still perform the core work; AI helps them do it more efficiently. This creates incremental gains, often meaningful ones, but the underlying ratio of staff to beneficiaries remains relatively constant. An AI-native organization designs its programs around AI from the beginning, which means the ratio of staff to people served can be orders of magnitude different.

At Jacaranda Health, a clinical nurse does not personally respond to every one of the 10,000 daily health questions the PROMPTS platform receives. The AI system handles approximately 70% of those questions autonomously, triaging and routing the remaining 30% to nurses based on risk level. The nurses' expertise is directed toward the cases that genuinely require human clinical judgment. This is a fundamentally different organizational design than one where nurses handle questions one at a time.

Jacaranda Health's challenge was in some ways even more demanding than Tarjimly's. Maternal mortality in Kenya and similar settings is not primarily a resource problem in the sense of lacking hospitals or medications. It is an information problem: women and families do not receive timely, personalized guidance that helps them recognize danger signs and seek care before complications become fatal. Solving that problem through traditional means would require an army of community health workers maintaining ongoing relationships with millions of pregnant women across geographically dispersed communities.

The PROMPTS platform (Promoting Mothers in Pregnancy and Postpartum Through SMS) addresses this gap by treating AI not as a supplementary tool but as the primary interface between the health system and individual mothers. Each woman enrolled in PROMPTS receives personalized two-way messaging based on her specific medical history, location, gestational age, and socioeconomic circumstances. The system knows, for instance, if a woman has a history of preeclampsia, and its messaging during her third trimester reflects that elevated risk profile.

The technical architecture that makes this possible relies on a large language model trained on more than 1 million real health questions and answers in Swahili, built on Meta's open-source Llama 3 and hosted on AWS infrastructure. When a mother sends a message asking whether her headache requires a hospital visit, the AI has been trained to recognize the difference between a tension headache that warrants reassurance and a headache that, combined with other symptoms and her risk profile, should trigger an urgent escalation to a clinical nurse. The system handles approximately 70% of the 10,000 to 12,000 daily inquiries autonomously; the remaining 30% are routed to nurses with contextual information that helps them respond efficiently.

The outcomes evidence is striking. A published cluster randomized controlled trial confirmed a 20% increase in prenatal health visit attendance and a 1.85x increase in postpartum family planning adoption among PROMPTS participants. In counties participating in the Kenya Quality Ecosystem coalition, facility-based maternal mortality declined by 29%. Makueni county increased its maternal and newborn health budget by 45% after seeing the data, a sign that AI-enabled impact measurement was changing political will as well as health outcomes.

Jacaranda Health has served 3.8 million mothers in Kenya and is actively expanding to Ghana, Nigeria, and additional countries, with a partnership with eHealth Africa targeting 25,000 mothers in Nigeria. As Jay Patel, Jacaranda's Director of Technology, described it: "AI helped us shift human intelligence to the urgent clinical questions, so we can quickly direct parents with critical needs to the right care." That shift, from using humans for volume to using AI for volume and humans for judgment, is the defining feature of the AI-native model.

Most organizations reading about Tarjimly and Jacaranda Health cannot simply rebuild themselves from the ground up as AI-native entities. They have existing staff, funders, programs, and relationships built over years or decades. The question for established nonprofits is not how to become AI-native overnight, but how to apply AI-native thinking to their own work.

The most transferable lesson from AI-native organizations is the discipline of asking, for every function and workflow: what portion of this work requires human judgment, and what portion is essentially a coordination, routing, triage, or information-retrieval task? The coordination and routing work is almost always where AI can create the most leverage. In a traditional nonprofit, this work is often done by experienced staff who are consequently unavailable for the higher-value relationship and judgment work that only humans can do well.

Donor relations provides a useful example. A development team at a mid-sized nonprofit might spend considerable staff time on tasks like identifying which donors have not been contacted recently, drafting routine stewardship updates, researching donor interests before calls, and scheduling follow-ups. These tasks are important but not inherently difficult. AI tools can handle much of this coordination work, freeing development staff for the relationship conversations that actually drive major gifts. Organizations that have made this shift consistently report that gift officers feel less burned out and more effective, because they are spending more time doing the work they find meaningful. For more on this approach, see our piece on virtual engagement officers.

Program delivery offers a similar opportunity. Organizations that provide case management, counseling referrals, or information services often have staff spending significant time on intake assessments, eligibility screening, and information-gathering before a client reaches a human professional. AI systems can handle much of this front-end work, with human staff stepping in when clients have complex needs or when AI triage flags elevated risk. This is precisely the model Jacaranda Health uses, and it is applicable in social services, legal aid, mental health intake, and many other direct service contexts.

One practical barrier for established nonprofits seeking to adopt AI-native approaches is funding. Building the kind of sophisticated AI infrastructure Tarjimly or Jacaranda Health relies on requires upfront investment that most traditional operating budgets cannot accommodate. Tarjimly's "First Pass" tool was built with a $1.3 million Google.org grant specifically designated for technology development. Jacaranda Health built its AWS-hosted AI system with technology grants and impact investors.

The funding landscape for nonprofit AI infrastructure has shifted meaningfully in the past two years. Major technology companies including Google, Microsoft, and Amazon have significantly expanded their nonprofit technology grant programs, and foundations focused on social impact are increasingly receptive to proposals that frame AI investment as a multiplier of program effectiveness. The key framing that resonates with funders is not "we want to use AI" but "we want to serve twice as many people with the same program budget, and this is how AI enables that."

For organizations that cannot access large technology grants, the path to AI-native thinking does not require building custom systems. Commercial AI tools available at nonprofit pricing through TechSoup and similar programs offer many of the coordination and triage capabilities that AI-native organizations use. The investment required is not always in technology; often it is in organizational design, clear definitions of which tasks belong to AI and which belong to humans, and the ongoing attention needed to ensure AI systems are performing as intended. Organizations considering their overall AI investment strategy can find guidance in our piece on justifying AI investment to your board.