AI in health financing: Asia-Pacific policymakers identify key enablers and call for regional collaboration

Policymakers and health financing specialists from across Asia and the Pacific gathered at the ADB-WHO Forum on Harnessing AI for Health Equity in Manila to share practical experience of deploying artificial intelligence in health insurance and financing operations — and to identify the conditions that determine whether such efforts succeed or fail.

The session brought together representatives from China, Indonesia, the Philippines, and Uzbekistan, alongside specialists from GIZ’s openIMIS initiative and the Asian Development Bank.

Three enablers, consistently identified

Despite the diversity of national contexts represented, panellists converged on a remarkably consistent set of prerequisites for effective AI adoption in health financing.

Saurav Bhattarai, Team Lead at openIMIS, GIZ, argued that well-documented business processes are the starting point — without clarity on how operations such as claims submission actually flow, it is impossible to identify where AI can add value.

He illustrated the importance of data consistency with a case from Nepal, where an AI-based claims adjudication model suffered a sharp and initially inexplicable drop in performance that was ultimately traced to COVID-19 lockdowns disrupting the working patterns of the human reviewers whose decisions had been used to train it.

Setiaji Setiaji, Director of Information Technology at BPJS Kesehatan — Indonesia’s national health insurance body, which manages coverage for more than 281 million people and processes over 1.4 million transactions daily — described data integrity and interoperability as non-negotiable foundations.

He also stressed the importance of explainable AI, particularly in claims adjudication: a system that rejects claims without clear reasoning cannot build the trust of healthcare providers. A third enabler he highlighted was the regulatory sandbox — a policy environment allowing new tools such as fraud detection modules to be tested at limited scale before full deployment.

Francis Uy, CEO of Sinag Solutions in the Philippines and a WHO digital health architecture adviser, framed AI not as a standalone intervention but as one component of a broader digital health transformation.

He identified five dimensions that must all be addressed in any credible pilot: clarity of goals, business architecture, organisational readiness, data readiness, and technology readiness — cautioning that neglecting any one of them undermines the prospects for scale.

China’s approach: cloud imaging and open competition

Zhang Yuanzhi, Director at China’s National Healthcare Security Administration, outlined two areas of focus for NHSA. The first is a unified cloud system for medical imaging, which allows patients to share scan histories across hospitals and provinces, eliminating repeat imaging and the associated costs.

The second is an AI competition launched the previous week in Guangxi province, focused on cancer detection from medical images, with more than 1,000 teams expected to participate from leading companies, universities, and hospitals. Zhang extended an open invitation for teams from across the Asia-Pacific region to enter the global competition and attend the final in China in October.

He also highlighted the broader equity argument for AI in medical imaging: top radiologists are concentrated in major cities, and AI can extend diagnostic accuracy to rural patients who would otherwise face long travel or risk of misdiagnosis.

Indonesia’s edge computing approach

With more than 10,000 islands and significant connectivity gaps, Indonesia’s BPJS Kesehatan has developed AI models capable of running offline on low-memory devices — tablets and laptops without GPU requirements — to serve remote facilities.

Setiaji described using biometric fingerprint recognition integrated with the national identity database to verify patients at the point of care even in areas with limited bandwidth. Fraud detection models trained on hundreds of millions of local claims records have achieved accuracy rates of around 87 percent.

On satellite connectivity, Setiaji confirmed that Indonesia has deployed Starlink access to more than 300 remote primary care facilities as part of efforts to address infrastructure gaps — a figure that drew a response from Junghwan Park of South Korea’s Ministry of Health and Welfare, who noted that satellite connectivity had also become an increasingly affordable solution for rural areas in Korea.

Cross-country collaboration: building a regional community of practice

A question from Junghwan Park of South Korea’s Ministry of Health and Welfare prompted broader discussion about whether Asia’s health data — spanning enormous and diverse populations — could be harnessed at a regional rather than purely national level.

Uy suggested that existing regional bodies such as the Asia eHealth Information Network could serve as vehicles for knowledge-sharing and collaborative learning, while noting that data sovereignty and legal frameworks would need careful attention.

He cited the Hajj pilgrimage health data system — through which participating countries share patient records to support care during the pilgrimage in Saudi Arabia — as a precedent for cross-border health data collaboration when the value proposition is clear and the governance framework is in place.

Dr Alvin Marcelo of the Asia eHealth Information Network proposed post-conference collaboration between the network’s eleven country laboratories, the WHO reference architecture framework, and China’s AI competition — suggesting the forum itself had created conditions for concrete follow-up.

Shared caution on imported models

Several panellists cautioned against deploying AI models developed in other contexts without local adaptation. Setiaji noted that models trained on foreign disease profiles and social determinants of health may perform poorly in Indonesia, and described a deliberate strategy of using international foundation models as a starting point before retraining on local data.

Bhattarai echoed this from the openIMIS experience, noting that the open-source approach gives lower-income countries a head start without the full investment of building models from scratch, provided local training data of sufficient quality is available.