Health systems across Asia-Pacific have adopted generative AI faster than they have built the infrastructure to support it, and the gap is showing up in stalled pilots and exceeded budgets. A HIMSS 2026 survey of AI implementation in APAC healthcare found 81% of respondents reported using generative AI, yet only 35% had a dedicated internal AI team, and 67% named cost as the most cited barrier to broader adoption.
The cost pressure lands hardest where the groundwork is thinnest. Decades of laboratory results, imaging reports, medication histories and clinical notes sit in systems never designed to interoperate, in varying formats and often in different languages. One institution Jonda Health worked with held a document archive exceeding three million files, with valuable clinical data accessible only to someone reading a single record, not to a system querying millions.
Consolidation is not readiness
Pulling records into a data lake is a necessary foundation, but it is not a destination. The teams Jonda Health works with rarely describe a shortage of data, said founder Suhina Singh, whose clinical practice experience showed her the cost of broken data: a value captured at one hospital could not be reliably compared with the same value from another. "The information she needed was often technically present but practically out of reach," her team said.
Harmonisation changes that. A laboratory trend becomes readable across a decade and multiple institutions; a medication history from another country's system becomes understandable; a discharge summary in one language works in another. Jonda Health does this by cleaning records in whatever format and language they arrive in, resolving results captured a dozen different ways, quality-assuring output with clinical review, then mapping data to consistent medical coding standards.
Singh's experience of the mismatch between what AI promises and what data systems deliver is common. "The teams we work with rarely describe a shortage of data; they describe data that cannot yet be trusted to mean the same thing twice."
AI is the right tool, but not the cheap shortcut
AI is itself the tool for this work - Jonda Health uses large language models to clean and map records. But teams that point a general-purpose model at millions of records quickly discover the cost of token-inefficient pipelines, and the harder problems surface later: identifiable patient data passing through models in ways that were never assessed for compliance, and output entering clinical or research use without the checks and balances needed to catch a wrongly extracted value.
The choice of where to begin, the team says, is not whether to do this work but where to start. This is why the team advises starting with a high-volume data type or a reporting use case that will show concrete results quickly. A data lake can look full and still return nothing.
The economics of healthcare AI feel opaque partly because the models attract attention while the real cost sits underneath - cleaning, structuring and standardising data. "The true scale of that effort is rarely visible to the people budgeting for it," Singh said.
The APAC data advantage
There is also a potential asset side. Well-structured clinical data has become one of the scarcest inputs in medical research. Drug developers, research organisations and companies building clinical AI depend on it, and the constraint is almost never the volume of data - it is the state of that data. Data becomes usable and valuable only once it has been harmonised, quality-assured and rigorously de-identified and pseudonymised.
Asia-Pacific institutions hold a version of this asset that the rest of the world is short of. An analysis in JACC: Asia examined 656 cardiometabolic trials published between 2011 and 2020 and found only 8.3% of participants were Asian - despite the region being home to roughly 60% of the world's population and, by the authors' own account, around 60% of the global burden of the diseases studied. The multilingual, ethnically diverse, longitudinal records that health systems here have struggled to manage are precisely the records global research most needs.
In the United States and parts of Europe, institutions have begun licensing carefully de-identified data to research partners. Most leaders across APAC do not doubt their data has value. The harder question is how to turn into an asset that strengthens the institution's own clinical work while, in a regulated way, becoming inspectable a tradeable asset in its own right.
Why this matters for healthcare professionals
For clinicians and administrators on the ground, the nature of the work stays the same, but the timeline changes. A health system that prioritises data readiness and a continuous capability, not a one-off migration, will be the one that holds an advantage that is hard to replicate later.
"The health systems that come through this period with a strong patient view and a strong data asset will not necessarily be the ones with the largest models," Singh said. "They will be the ones that turn their chaotic, historical records into clarity, and meet the privacy bar with discipline."
This is also where the broader question of data value comes in - and where the line becomes crucial. Rigorously pseudonymised, harmonised data can serve research without compromising the individuals behind the records. De-identification, pseudonymisation and anonymisation describe different protection levels, and the privacy bar exists for reason. "An institution that cannot demonstrate that discipline has nothing to trade," said the team.
For professionals working with health data day-to-day, the practical takeaway is that starting with a narrow, high-return use case - a specific clinical workflow, a research question, or clean incoming data effective for the whole organisation - beats a two-year promise of everything. A programme that shows a concrete result in its first months earns the room to keep going. A data lake that looks full might still return nothing.
Your work is within your reach - and if you're looking for ways to build these skills, there are pathways to AI for Healthcare Courses or specialised skill development for roles that handle AI for Medical Records Clerks. The foundation of your professional practice is being built now - by you. The records you've spent years managing are not only an operational cost but the raw material from which the next advances in medical knowledge will be made.
Your membership also unlocks:
