The last mile of health data and the first mile of AI

• by Sakshi Agarwal, Anirooddha Mukherjee and Dr. Puneet Khanduja
• Sep 21, 2026
• 4 min

This blog explores how fragmented health data systems in India can limit artificial intelligence (AI) effectiveness in healthcare. It highlights the need to reduce repetitive data entry, redesign frontline workflows, strengthen interoperability, and build connected, longitudinal systems for reliable AI solutions.

In a rural health catchment in northern India, an Auxiliary Nurse Midwife (ANM) worker updates a pregnant woman’s details on the Reproductive and Child Health (RCH) portal after a household visit. Later that day, the same information must be entered again into other systems for facility reporting and program tracking. Hours that could be spent on follow-up care are instead spent across platforms that were never designed to work together. 

This operational reality rarely comes up in conversations about artificial intelligence (AI) in healthcare with an adequate and workable solution. Yet it is precisely at the point of data generation and entry that the effectiveness of future AI systems takes shape. The quality of the data that enters these systems, and the way it moves between them, influences what AI can do later. 

India has invested heavily in digital health infrastructure. Across maternal and child health, tuberculosis (TB), human immunodeficiency virus (HIV), immunization, and disease surveillance, these platforms capture vast amounts of data every day. At the same time, there is growing momentum to deploy AI, from clinical decision support to AI-assisted diagnostics.  

Yet, AI systems are only as reliable as the data environments they are built on. For an Accredited Social Health Activist (ASHA), the information entered during a household visit does not necessarily remain in one place. A pregnant woman’s information may be recorded separately across maternal tracking, facility reporting, and immunization workflows. Small inconsistencies, such as spelling variations, incomplete histories, and missing identifiers, make it hard to build a longitudinal picture of care. AI does not make these gaps disappear, and it often makes them harder to detect. The real challenge lies in the fragmented structure of the systems that AI tools enter, rather than the absence of AI tools themselves.  

The structure shapes the outcome 

India’s health data architecture evolved program by program: HMIS for facility reporting, the RCH portal for maternal and child tracking, Ni-kshay for TB management, and IHIP for disease surveillance. Each system was built independently to meet specific operational needs. The result is parallel platforms with distinct workflows, identifiers, and requirements. More recently, the JANANI platform has sought to address some of these gaps by consolidating maternal and child records into a single longitudinal system, though its rollout is still at an early stage and will take time to stabilize. 

A patient’s TB history, maternal records, immunization status, and chronic conditions may sit in disconnected systems, which limits an integrated view of care across conditions. This is more than a technical issue. This reflects the way public health programs evolved in institutional silos, and AI systems built on top inherit those same gaps. 

Why this matters for AI 

AI learns patterns from the data it receives, but fragmented systems often provide only partial snapshots rather than a connected continuum of care. A predictive model may miss a high-risk patient because the system around it never captured an overview of that patient’s journey, even when the algorithm itself is sound. AI outputs can appear precise and credible while relying on incomplete data. As a result, AI across fragmented ecosystems may reinforce blind spots rather than resolve them. 

Why interoperability alone may not be enough 

Most efforts to date have focused on interoperability, which enables systems to exchange data. India’s Ayushman Bharat Digital Mission has invested in shared standards, digital IDs, and registries. These are necessary foundations; however,  they do not change how data is generated on the ground.  

Frontline workers still navigate multiple platforms, and they enter the same information repeatedly because systems stay organized around separate mandates rather than a unified patient journey.  

Interoperability helps systems talk to each other after the fact, but it does not stop health workers from entering the same data repeatedly at the source. Fragmentation begins at the point of data capture itself. 

The missing layer 

Between frontline data collection and system-level intelligence lies a less-discussed layer that determines how information moves across the continuum of care. 

Today’s platforms reflect program boundaries, rather than care continuity, so integration often happens only after data has already entered the ecosystem in fragmented form. Backend fixes alone cannot always fully address these issues later. 

A better approach may be to redesign workflows at the point of entry so that frontline workers capture information once through shared interfaces. This data can then flow into program-specific systems as needed. This would not eliminate program-specific requirements but could reduce duplication and build a stronger foundation for AI that depends on connected, longitudinal data.  

For the ASHA worker, reducing the burden of repeated data entry can create more space for work beyond the screen, such as patient care, follow-up, and community engagement. MSC is testing this approach in Bihar, where ASHA incentive claims benefit from streamlined processes through the removal of redundant data entry, and in Odisha, where system redesign supports improvements in maternal and child health indicators.  

Through its work with ICMR–National Institute for Research in Digital Health and Data Science (ICMR-NIRDHDS), MSC supports the development of national guidance instruments that directly address the governance dimensions of this challenge. These instruments include frameworks for data curation, data sharing, and interoperability standards for AI systems in public health. A multi-stakeholder Technical Working Group convenes this work and translates the structural data challenges described here into actionable policy direction at the national level. 

India has built strong digital infrastructure and now seeks to add intelligence on top. Yet, intelligence is only as coherent as the systems beneath it.  

The real question may be: can healthcare AI ever be truly integrated if the health system itself still operates in silos?