India doesn’t have a hospital AI problem. It has a data problem

Last reviewed 23 September 2026

A new Bain and HealthQuad report says most Indian healthcare providers are still running AI pilots, with real scale limited to operational uses and electronic medical record adoption at around 35%. The bottleneck is not the models. It is the data the models need.

What happened

The report, summarised by IndianWeb2, finds that AI in Indian healthcare is scaling first in operational and workflow areas, while clinical use is mostly limited to mature providers using it as decision support. It puts EMR adoption at roughly 35%, concentrated in large urban chains, and notes that many small and mid-sized hospitals still rely on paper records.

The blockers it names are data readiness, regulatory clarity and talent, with deep AI skills often pointed at global markets rather than Indian providers. It also points out that model costs have fallen sharply since 2023, which makes the data gap even more obvious. The tools are getting cheaper. The inputs are not getting better at the same pace.

My take

When I hear a hospital leader say “we want to do AI”, the most useful follow-up question is “on what data?” In most Indian hospitals the honest answer is uneven. The clinical record might be partly digital, partly scanned, partly in a doctor’s handwriting. Lab and imaging systems may not talk to each other. Discharge summaries vary by department.

That is why so many pilots stall. A clinical AI model that works in a demo needs clean, structured, longitudinal data to work in production, and that data often does not exist yet. The pilot ends, the vendor moves on, and the board concludes that AI is overhyped.

Where the data already is

There is a part of the hospital where data is already digital, structured and plentiful: the patient front door.

• WhatsApp and chat. Every enquiry, question and booking request is already text.
• The contact centre. Calls are recorded, tagged and increasingly transcribed.
• Appointments and payments. Booking systems and billing already produce clean event data.
• Reviews and feedback. Patients tell you, in writing, what went wrong.

This is where AI can go live quickly and safely. Routing enquiries to the right specialty, answering routine questions about timings and preparation, summarising calls, flagging unhappy patients before they post a review, predicting no-shows: none of this touches a diagnosis, and all of it has data ready today.

What most coverage missed

Front-door AI is also how you fix the clinical data problem over time. Every structured intake conversation is a small piece of clean data. Every digitised report a patient uploads before a visit reduces paper. If the patient-facing layer is designed well, it becomes the funnel through which the hospital’s data quality improves, one interaction at a time.

It also builds the internal trust that clinical AI will need later. A team that has seen an assistant handle ten thousand booking conversations safely is far more open to a documentation tool in the consulting room than a team whose only experience is a failed pilot.

What I would do

• Pick two front-door use cases with clear measures, such as enquiry response time and appointment conversion.
• Make structured data capture a design requirement for every digital channel, not an afterthought.
• Set a data readiness score for each clinical AI idea before approving a pilot.
• Keep a human review line for anything that sounds like medical advice.

India will get to clinical AI at scale. The route there runs through the channels patients already use, and through the unglamorous work of getting data right.

Go deeper: The AI strategy a hospital group can actually execute, Measuring an AI project when nothing clinical is being claimed and The pilot that has to work before automation scales.

Source: IndianWeb2 (Bain and HealthQuad report). Figures as reported at the time of writing.

Questions people ask