{"id":46256,"date":"2026-03-31T09:13:22","date_gmt":"2026-03-31T09:13:22","guid":{"rendered":"https:\/\/www.microsave.net\/?p=46256"},"modified":"2026-03-31T09:15:05","modified_gmt":"2026-03-31T09:15:05","slug":"the-missing-half-of-digital-health-exploring-pathways-for-ai-to-scale-in-lmic-health-systems","status":"publish","type":"post","link":"https:\/\/www.microsave.net\/2026\/03\/31\/the-missing-half-of-digital-health-exploring-pathways-for-ai-to-scale-in-lmic-health-systems\/","title":{"rendered":"The missing half of digital health: Exploring pathways for AI to scale in LMIC health systems"},"content":{"rendered":"

Over the past decade, health ministries\u00a0in\u00a0many\u00a0low- and moderate-income\u00a0countries\u00a0(LMICs)\u00a0have experimented with a new generation of AI tools in everyday healthcare settings. In Pakistan, India, and parts of\u00a0Sub-Saharan Africa, AI systems\u00a0are being\u00a0tested to\u00a0<\/span>help screen patients<\/span><\/a>\u00a0for tuberculosis.\u00a0Hospitals in Kenya and Bangladesh have\u00a0begun\u00a0testing algorithms\u00a0that flag pregnancies at higher risk of complications.\u00a0In Nigeria, primary health centers\u00a0have piloted digital tools that\u00a0<\/span>offer guidance<\/span><\/a>\u00a0during clinical decisions. Many of these experiments have produced promising results,\u00a0which\u00a0prove\u00a0that\u00a0the technologies\u00a0work.\u00a0Yet\u00a0very few have moved beyond the pilot stage<\/span>,\u00a0<\/span>a fate similar to\u00a0<\/span>digital health tools<\/span><\/a>.<\/span>\u00a0<\/span><\/p>\n

The\u00a0challenge\u00a0has\u00a0been\u00a0the system\u00a0rather than\u00a0the technology, a reality the global health community has been slow to acknowledge.\u00a0This\u00a0phenomenon, described in\u00a0the\u00a0digital health\u00a0literature\u00a0as\u00a0<\/span>pilotitis<\/span><\/a>,\u00a0is widespread across LMIC health systems.\u00a0Pilots are\u00a0plentiful,\u00a0but\u00a0integration\u00a0into existing health systems\u00a0is rare.<\/span>\u00a0<\/span>Many innovations\u00a0show\u00a0strong results under short-term donor funding, only to collapse once that support ends.\u00a0The technology itself may remain\u00a0viable, but for lasting impact, server space continuity, re-skilling of staff, and financial planning are important pre-requisites.<\/span>\u00a0<\/span><\/p>\n

This\u00a0gap between promising pilots and\u00a0real-world scale is not unique to health systems. Across industries, most AI initiatives\u00a0<\/span>struggle<\/span><\/a>\u00a0to move beyond\u00a0proof\u00a0of\u00a0concept.\u00a0Industry\u00a0estimates\u00a0indicate\u00a0that\u00a0<\/span>nearly a third of generative AI projects<\/span><\/a>\u00a0may be abandoned after the pilot stage due to unclear value or implementation barriers.<\/span>\u00a0<\/span><\/p>\n

The uncomfortable truth is that most\u00a0of the\u00a0scaling frameworks applied in LMICs were designed for high-income health systems with different infrastructure, institutional maturity, and health worker profiles.\u00a0These frameworks often fall short\u00a0in LMIC settings,\u00a0which highlights\u00a0the need to\u00a0develop approaches\u00a0customized\u00a0to\u00a0their own realities.<\/span>\u00a0<\/span><\/p>\n

Why the standard playbook fails<\/span><\/b>\u00a0<\/span><\/p>\n

In practice, the\u00a0dominant\u00a0approach\u00a0to AI in health\u00a0has been\u00a0to build\u00a0technology\u00a0outside the systems it is meant to\u00a0serve.\u00a0They\u00a0are\u00a0built\u00a0generally in\u00a0high income countries,\u00a0tested in donor-funded pilots,\u00a0and\u00a0validated\u00a0under\u00a0controlled conditions. They are\u00a0then handed to\u00a0a public health system\u00a0that\u00a0had\u00a0no\u00a0meaningful involvement\u00a0in\u00a0their\u00a0development. This model has\u00a0consistently struggled\u00a0at the\u00a0scaling stage, and the\u00a0<\/span>reasons are\u00a0well-documented<\/span><\/a>.\u00a0The\u00a0<\/span>IBM Watson for Oncology<\/span><\/a>\u00a0<\/span>is\u00a0a pertinent example. Trained on data from a US cancer center\u00a0and deployed across hospitals in India, Thailand, and South Korea,\u00a0it\u00a0collapsed in\u00a0practice\u00a0because\u00a0its recommendations were clinically incompatible with local health systems that\u00a0had no role in\u00a0shaping\u00a0it.<\/span>\u00a0<\/span><\/p>\n

Several systemic constraints limit the ability of AI tools to scale within LMIC health systems.\u00a0The first constraint is\u00a0<\/span>connectivity<\/span><\/b>.\u00a0<\/span>Many AI tools assume reliable broadband and stable digital infrastructure. These\u00a0conditions are often absent in rural areas,\u00a0where much of the disease burden in LMICs is concentrated.\u00a0One example is the\u00a0real-world deployment of\u00a0<\/span>an artificial intelligence tool for chest X-rays.<\/span><\/a>\u00a0Recurring\u00a0incompatibilities between\u00a0the AI software and existing X-ray equipment were documented due to differences in\u00a0image formats and acquisition protocols.<\/span>\u00a0<\/span><\/p>\n

The second constraint is\u00a0the\u00a0<\/span>digital maturity of health workers<\/span><\/b>.\u00a0It\u00a0receives the least investment and\u00a0has the\u00a0most optimistic assumptions. Frontline workers are often\u00a0highly skilled at navigating community\u00a0health\u00a0and adapting clinical\u00a0practice\u00a0under resource pressure.\u00a0However,\u00a0their experience\u00a0with digital tools is recent, uneven, and layered on top of workflows that have not changed significantly since the paper\u00a0register. An AI system that requires new data entry habits or\u00a0needs\u00a0health workers to\u00a0interpret\u00a0model outputs without sustained support\u00a0<\/span>will not be used consistently<\/span><\/a>, regardless of performance.<\/span>\u00a0<\/span><\/p>\n

The third constraint is\u00a0<\/span>data<\/span><\/b>. LMIC health information systems\u00a0are\u00a0fragmented,\u00a0mostly\u00a0built to serve\u00a0government\u00a0reporting cycles rather than clinical decision-making. An AI tool trained on data from Nairobi\u00a0<\/span>will not perform<\/span><\/a>\u00a0with the same accuracy in rural Kenya, let alone in a different country entirely.<\/span>\u00a0<\/span><\/p>\n

The fourth, and most structurally dangerous, constraint is\u00a0<\/span>funding design<\/span><\/b>. Most pilots\u00a0operate\u00a0within short-term donor funding cycles. Scaling a digital health intervention into\u00a0the\u00a0national infrastructure requires sustained investment\u00a0for\u00a0many years.\u00a0When funding\u00a0falls short,\u00a0<\/span>the intervention\u00a0stalls<\/span><\/a>\u00a0before it can\u00a0reach scale.<\/span>\u00a0<\/span><\/p>\n

Workable\u00a0pathways<\/span><\/b>\u00a0<\/span><\/p>\n

The\u00a0<\/span>first pathway<\/span><\/b>\u00a0is to design\u00a0specifically\u00a0for the constraints that LMICs face. AI\u00a0tools must\u00a0function offline,\u00a0operate\u00a0under low-bandwidth conditions, and work across a wide range of devices.\u00a0For example, computational power\u00a0(commonly referred to as\u00a0compute)\u00a0is a binding constraint that rarely\u00a0features in\u00a0pilot design.\u00a0Cloud-based AI inference requires reliable internet and ongoing server costs that few LMICs\u00a0can sustain at scale.\u00a0High-income health systems\u00a0often treat the\u00a0compute\u00a0infrastructure,\u00a0such as\u00a0stable cloud access, sufficient processing power,\u00a0and\u00a0manageable API costs,\u00a0as\u00a0background conditions.\u00a0<\/span><\/p>\n

Systems\u00a0must actively plan\u00a0their AI tools\u00a0for\u00a0use in\u00a0low-resource settings.\u00a0One\u00a0option\u00a0is the\u00a0use\u00a0of\u00a0<\/span>Small Language Models<\/span><\/a>, or\u00a0\u201c<\/span>small AI<\/span><\/a>\u201d, which is\u00a0lightweight, locally deployable models\u00a0that run\u00a0on low-cost devices without cloud connectivity.\u00a0Large-scale AI systems\u00a0may\u00a0deepen existing inequities\u00a0by\u00a0perpetuating bias\u00a0due to\u00a0underrepresented\u00a0LMIC populations in their training data. Small AI\u00a0offers a more\u00a0viable\u00a0and contextually\u00a0appropriate\u00a0alternative. The\u00a0TB detection tools with the\u00a0greatest durability at scale ran on laptops\u00a0<\/span>without internet access<\/span><\/a>\u00a0and integrated with mobile X-ray units in remote communities.\u00a0These\u00a0include\u00a0CAD4TB deployments in Pakistan and Bangladesh.\u00a0Features,\u00a0such as voice interfaces for workers with limited text literacy and outputs aligned with\u00a0decisions made by\u00a0community health workers,\u00a0were found to be\u00a0contextually\u00a0superior.\u00a0<\/span>\u00a0<\/span><\/p>\n

Task-specific small language models (SLMs) trained on locally representative clinical data and deployable on low-cost hardware represent one of the most underleveraged opportunities in Health AI for LMICs. Unlike general-purpose models that depend on cloud connectivity and commercial APIs, sovereign SLMs for narrow use cases like TB X-ray interpretation or high-risk pregnancy detection can outperform larger models on the tasks that matter, while keeping data governance and long-term costs firmly in government hands.<\/span>\u00a0<\/span><\/p>\n

Based on this agenda, MSC\u00a0(MicroSave\u00a0Consulting)\u00a0cofounded\u00a0<\/span>the Alliance for Inclusive AI<\/span><\/a>\u00a0with\u00a0<\/span>BFA Global<\/span><\/a>\u00a0and\u00a0<\/span>Caribou<\/span><\/a>. We are committed to developing practical \u201csmall AI\u201d solutions that expand\u00a0opportunity\u00a0for underserved communities across the Global South.<\/span>\u00a0<\/span><\/p>\n

The\u00a0<\/span>second pathway<\/span><\/b>\u00a0prioritizes the digital readiness of\u00a0health workers, way before deployment begins.\u00a0This begins with an honest assessment\u00a0of baseline digital capacity before implementation,\u00a0led by health ministries.\u00a0This would entail a phased rollout starting in facilities with higher digital literacy, generating evidence and frontline champions, then expanding to the periphery with those champions as peer trainers.\u00a0\u00a0Thus, frontline workers\u00a0would actively help design tools.<\/span>\u00a0<\/span><\/p>\n

The\u00a0<\/span>third pathway<\/span><\/b>\u00a0is data governance from the outset. Governments\u00a0that\u00a0launch\u00a0AI pilots must\u00a0establish\u00a0data ownership, consent frameworks, and interoperability standards before deployment. These foundations must be in place\u00a0before\u00a0the pilot has generated a dataset that becomes proprietary to an external developer. India’s\u00a0<\/span>Ayushman Bharat Digital Mission<\/span><\/a>,\u00a0even as it continues to mature operationally,\u00a0represents\u00a0a serious effort\u00a0to build the data architecture that AI-enabled health systems require.\u00a0The lesson for other LMICs is to invest in this infrastructure before the AI tools arrive.<\/span>\u00a0<\/span><\/p>\n

The\u00a0<\/span>fourth pathway<\/span><\/b>\u00a0is to\u00a0plan the\u00a0government\u00a0financing\u00a0transition\u00a0early.\u00a0Every pilot should include a clear and costed pathway for integration into public\u00a0budgets well\u00a0before donor funding expires.\u00a0Governments\u00a0must\u00a0engage\u00a0finance ministries early,\u00a0show\u00a0value\u00a0through the\u00a0cost per\u00a0<\/span>disability-adjusted\u00a0life\u00a0year<\/span><\/a>\u00a0(DALY)\u00a0averted, and negotiate\u00a0innovation-friendly\u00a0procurement frameworks, especially at provincial or state levels.\u00a0When\u00a0governments plan this transition early, AI-enabled tools offer the potential to reduce cost per patient reached and strengthen the evidence base for more informed healthcare budget allocations at national and sub-national levels.<\/span>\u00a0<\/span><\/p>\n

The\u00a0<\/span>fifth pathway<\/span><\/b>\u00a0is\u00a0a\u00a0regulatory architecture that enables\u00a0responsible\u00a0and\u00a0timely\u00a0deployment.\u00a0Many LMIC ministries\u00a0lack\u00a0comprehensive AI-specific regulatory frameworks. In such cases,\u00a0waiting for a fully developed\u00a0framework\u00a0can\u00a0stall\u00a0useful innovations.\u00a0Regulatory sandboxing, time-limited conditional approvals tied to post-market surveillance, and regional\u00a0regulatory\u00a0harmonization frameworks adapted from\u00a0<\/span>pharmaceutical precedents<\/span><\/a>\u00a0offer pragmatic\u00a0mechanisms\u00a0to\u00a0introduce\u00a0AI tools\u00a0and\u00a0maintain\u00a0safeguards.<\/span>\u00a0<\/span><\/p>\n

The\u00a0<\/span>sixth pathway<\/span><\/b>, and the most underused, is LMIC-to-LMIC learning. Contexts that share infrastructure constraints, workforce realities, and regulatory gaps are better positioned to learn from each other than from high-income systems where these conditions do not exist. Structured knowledge exchange, supported by regional bodies, offers a more grounded and transferable basis for scaling than imported playbooks\u00a0ever could.<\/span>\u00a0<\/span><\/p>\n

\"\"<\/span><\/p>\n

Figure\u00a01:\u00a0Six pathways for governments to move from health AI pilots to population-level impact\u00a0<\/strong><\/p>\n

The\u00a0way forward<\/span><\/b>\u00a0<\/span><\/p>\n

The global health community has\u00a0invested considerable effort\u00a0to\u00a0show\u00a0that AI tools can\u00a0operate\u00a0in low-resource settings.\u00a0Much of the\u00a0evidence is\u00a0promising, and technology\u00a0itself is not the main obstacle.\u00a0<\/span>\u00a0<\/span><\/p>\n

The challenge lies in the systems into which these tools must be introduced,\u00a0as\u00a0health systems are complex institutions\u00a0and\u00a0not controlled environments.\u00a0Scaling a technology requires infrastructure\u00a0to support it, workers who can use it confidently, data systems that can provide\u00a0reliable inputs, governance structures that can regulate it, and public budgets that can sustain it.\u00a0In\u00a0many\u00a0LMIC health systems,\u00a0these foundations must be\u00a0deliberately built.\u00a0<\/span>\u00a0<\/span><\/p>\n

This is a familiar lesson.\u00a0Expanding coverage without governance\u00a0produced\u00a0the paradox we see in health insurance systems across the Global South. Pilots without pathways will produce the same paradox in AI health systems.<\/span>\u00a0<\/span><\/p>\n

Success will therefore depend less on the number of promising pilot projects and more on the conditions that\u00a0enable scaling.<\/span>\u00a0<\/span>Governments that invest early in infrastructure, workforce readiness, data systems, and financing arrangements will be better positioned to move beyond experimentation.\u00a0<\/span>\u00a0<\/span><\/p>\n