By [Author Name] | Health & Technology Desk
A hospital can license a world-class diagnostic algorithm in an afternoon. Getting that algorithm to read a scan inside a working radiology department, on the hospital's own systems, without slowing anyone down — that is a different problem entirely.
At the sixth edition of the ET Healthcare Leaders Summit, three people who sit close to that problem said it out loud: the barrier to AI in Indian hospitals is not intelligence. It is integration.
The Diagnosis From the Summit Floor: The Model Isn't the Problem
The view came from three panellists who rarely sit on the same side of a healthcare debate. J.P. Dwivedi is Chief Information Officer at the Rajiv Gandhi Cancer Institute and Research Centre — a hospital-side technology leader. Dr. Saurav Basu is a Senior Scientist at the Indian Council of Medical Research (ICMR), the country's apex biomedical research body. Kalyan Sivasailam is Co-Founder of 5C Network, a private health-technology venture.
Public sector, private hospital and startup. Their shared conclusion, as reported: integration is the real barrier.
What "Integration" Actually Means Inside a Hospital
It is not a bureaucratic word. In practice, integration is everything that has to happen between buying an AI tool and a doctor trusting its output.
It means pulling patient data out of hospital information systems, radiology archives and laboratory records — systems often built by different vendors at different times, speaking different data formats. It means fitting an AI result into the clinician's existing workflow instead of asking them to open a second screen. It means deciding who is accountable when the model is wrong.
None of that is a machine-learning problem. All of it determines whether the model ever gets used.
Three Vantage Points, One Conclusion
Each speaker arrives at the same wall from a different direction — and that is what gives the observation weight.
A hospital CIO lives with the consequences of every integration decision: procurement cycles, legacy software, uptime, data security and staff who have to be trained on one more dashboard. A senior scientist at ICMR approaches adoption through evidence — whether a tool has been validated in Indian patient populations and clinical settings. A startup co-founder meets the same barrier from the outside, trying to deploy into hospitals that were never architected to accept outside software.
When all three describe the same obstacle, it is worth taking seriously.
Why a Working Demo Rarely Survives Contact With a Ward
Healthcare AI is usually demonstrated on clean, curated datasets. Hospitals run on messy, incomplete, inconsistently labelled records.
The gap between those two realities is where most pilots quietly end. A tool can be accurate in a study and still be unusable if it takes too long to load, cannot read the hospital's imaging format, or produces output no one has time to verify.
The panellists' framing suggests the sector has spent years optimising the wrong end of that chain.
Confirmed Facts vs What Remains Unclear
Confirmed: The panel discussion took place at the sixth ET Healthcare Leaders Summit. The three named participants hold the affiliations stated above. Their collective position was that integration, not the AI technology itself, is the principal barrier to adoption in hospitals.
Unclear: No specific deployment figures, cost estimates, hospital case studies, policy commitments or timelines were part of the account available at the time of writing. This report does not attribute statistics or quotations that were not disclosed.
This report is based on the stated conclusion of the session and the participants' professional affiliations. No independent high-confidence source on the full session was available at the time of publication.
The Moat Question: Why Integration Is Also a Business Story
If integration is the bottleneck, then the companies that solve it — not the ones with the best model — may end up owning the market.
That is a familiar pattern in health technology. Whoever connects to a hospital's data systems first becomes expensive to replace. Every additional hospital on the same network makes the product more useful and the switching cost higher. Clinical trust, regulatory comfort and existing procurement relationships compound the same way.
It also explains why AI vendors increasingly sell deployment and workflow support rather than algorithms alone. The algorithm is becoming a commodity; the plumbing is not.
Risks and the Balanced View
There is a counter-argument worth stating. Blaming integration can become an excuse for slow adoption — a convenient way for institutions to postpone decisions rather than make them.
There are also real dangers in integrating too fast: patient data flowing through poorly secured systems, unvalidated tools influencing clinical decisions, and clinicians deferring to software they do not fully understand. In a country with wide variation in hospital IT maturity, a rushed national push could widen the gap between well-resourced urban hospitals and everyone else.
The honest position sits between the two: integration is genuinely hard, and it is also not optional.
The Wider Pattern: A Digital Health Push Meets Hospital Reality
India has been steadily building a national digital health layer — health IDs, interoperable records, standardised data exchange. AI adoption is expected to sit on top of that foundation.
What the summit panel describes is the gap between the blueprint and the building. Standards exist on paper; hospital servers, vendor contracts and overstretched IT teams decide what actually gets deployed.
This is not unique to India, but the scale of the country's healthcare system makes the integration backlog larger and the payoff of solving it greater.
Practical Guidance: What Each Side Should Do Now
For hospital administrators: Treat integration cost as part of the AI budget, not an afterthought. Interoperability should be a question asked during procurement, not after installation.
For clinicians: Ask how a tool was validated, on whose data, and what happens when it is wrong. A tool you cannot interrogate is a tool you cannot safely rely on.
For health-tech founders: Design for the hospital you are actually deploying into — fragmented records, older software, overworked staff — rather than the clean environment of a pilot study.
For policymakers and researchers: Validation in Indian patient populations and shared integration standards may matter more to adoption than another funding round for model development.
Future Outlook
Expect the conversation to shift over the next few years from model accuracy to deployment reality — integration standards, procurement language, data-sharing frameworks and clinical validation in Indian settings.
Hospitals that solve integration early could pull ahead quickly. Those that do not may keep running pilots that never reach patients. No timeline or roadmap was announced at the session, and this remains an assessment rather than a forecast.
Our Take
The most useful thing about this panel was what it refused to do: sell AI. Instead, three people with very different incentives agreed on an unglamorous truth — the hard part of healthcare AI is not building it, it is wiring it into a system that was never designed for it.
That is a less exciting headline than a breakthrough model. It is also the one that decides whether patients ever see the benefit.
Frequently Asked Questions
What did the experts say is the real barrier to AI in hospitals?
Integration. Speaking at the sixth ET Healthcare Leaders Summit, the panellists said the difficulty of connecting AI tools to existing hospital systems — not the capability of the AI itself — is the main obstacle to adoption.
Who made this point at the summit?
J.P. Dwivedi, Chief Information Officer at the Rajiv Gandhi Cancer Institute and Research Centre; Dr. Saurav Basu, Senior Scientist at the Indian Council of Medical Research (ICMR); and Kalyan Sivasailam, Co-Founder of 5C Network.
Does this mean AI is not working in Indian hospitals?
No. The argument is about deployment, not capability. AI tools can perform well in controlled settings and still fail to reach patients if hospitals cannot integrate them into daily clinical workflows.
Why is integration so difficult in Indian hospitals?
Hospital records are typically spread across multiple systems built by different vendors, data formats vary, and IT teams are stretched. Adding an AI tool means connecting all of that, training staff and deciding who is accountable for its output — work that sits outside the AI model itself.
