Next Generation of Healthtech Should Be Won Behind the Scenes

AI changes the conversation about the future of healthcare. It has the potential to accelerate drug discovery, improve diagnostics, support quality, identify risk earlier and uncover patterns in data that we could not reasonably discover, we know this is the big selling point of AI.

This is especially powerful systemically to understand the causes of the causes and where neighbourhoods could invest limited resources. AI can help us see the inequity hidden across clinical, operational, social and economic data. It can help us understand deeper why some communities experience worse outcomes, where prevention is failing, and where the system itself creates delay, duplication or harm.

The next gen healthtech is then modelling the research outcomes for the hypothesis testing. The data tell us what happened yesterday, it can model behavior and new resources for tomorrow quicker and with small error margins through global learning, but the doing, that needs to be done, and is done by those inside healthcare. We are not machines that deploy what the machines tell us, the real world is more complex than this.

There is risk in the way some of the healthtech debate is developed. Too often, the conversation starts from a technocratic assumption that healthcare can be made cleaner, linear and more manageable if human interaction is shaped into the right digital process. Health and care is not a production line. It is not a neat sequence of tasks that can be packaged, standardised and automated. Healthcare is a complex set of human judgements, relationships, interruptions, risks, uncertainties, local knowledge and decisions made under pressure.

The purpose of healthtech should not be to force care into the logic of the computer. It should be to help people deliver care safely within the reality of healthcare. The tech is the tool, and like any tool it should make the job easier.

The distinction matters.

The next generation of healthtech will not be won only by the organisations with the most impressive AI demonstrations. It will be won by those who make the foundations work. The less glamorous things. The things behind the scenes. The ‘plumbing’ no one notices when it works, but everyone notices when it fails.

Interoperability is not a technical luxury. It is the clinical infrastructure. We have strong examples of what this direction can look like. Shared care records and regional information platforms can help professionals see more of the person and less of the organisational boundary, GNCR and SEER2. Done well, they support the reality of care, which is often delivered across hospitals, general practice, community services, mental health, social care, voluntary organisations and families.

The answer is not one large central system that everyone must adapt to. Centralisation creates consistency, scale and analytical power, but they can also miss local nuance. Care is delivered locally, even when it is planned or governed nationally. Local pathways matter. Local relationships matter. Local workarounds, pressures and clinical cultures matter.

A computer system that does not understand how care is actually delivered or how data flows will struggle, no matter how technically impressive it looks, or how much it costs. This is why many clinicians ask a simple and fair question: why can my phone do this, but the hospital computer can’t?

They can open an app, verify identity, move between services, receive notifications, share information and search quickly. Then they go to work with multiple logins, fragmented records, slow systems, missing information and digital processes that feel less intuitive than the technology they use in everyday life.

We know the reasons. Healthcare is complex. The stakes are high. Legacy systems are deeply embedded. Integration is expensive. Safety, privacy, procurement and governance all matter. Development costs money, implementation costs money, and doing it properly takes time. This is critical infrastructure. Do the people running power stations have the same questions?

But knowing why it is hard does not make the problem inevitably acceptable.

Health systems patch the problem. Then patch the patch. Then add a portal, a spreadsheet, a manual reconciliation process, a login, a dashboard, a workaround and another local process to bridge the gap.

Eventually, the staff find a workaround and this becomes the operating model. There is a reason we see whiteboards on ward rounds, admissions recorded on paper, and pen and clipboard processes running alongside expensive digital systems. It is not because staff are nostalgic for paper. It is because, in that moment, paper is easier, faster, then the computer.

That should make us and healthtech think. A whiteboard on a ward is not just a workaround. It is usability feedback.

The same applies to AI such as ambient voice. These tools have real promise. The goal should not simply be to help people navigate and record into the IT system. The goal should be to help people navigate healthcare. That means understanding the patient, the pathway, the local context, the organisational culture and the next safe action. It means AI should support the work of care, not simply feed the system more data by listening to us.

A good friend described the use of AI to me as being like a very new team member. They may have access to a great deal of knowledge, but they still need supervision, judgement, challenge and guidance in applying that knowledge safely. For me this a helpful way to think about AI in healthcare. It needs induction. It needs boundaries. It needs supervision. It needs escalation rules. It needs people who remain willing to challenge it, we must not let our first impressions influence our complacency, AI is probabilistic rather than deterministic, ie its ‘probably’ right, but we must be cautious, and I know that difficult when we’ve used it a while, it was good, and we are tired at the end of shift.

I am not glass-half-empty, there are good results seen in AI image analysis (for example) in mammograms, but this is work I saw ‘fully working’ in the early 2000s, long before its mainstream adoption. Timelines in healthcare are long and healthtech needs to be able to plan for this. This is not an anti-innovation mindset its the real assurance, adoption, and refinement of the tech.

The safety risk is not only that AI may be wrong or inaccurate. It is that it may be right often enough for people to stop asking whether it is wrong this particular time. That is automation complacency, and it should concern us, as it does affect us.

I have seen this when using large language models to support coding projects. They can be extremely useful. They can solve problems quickly, explain unfamiliar errors and accelerate development. They become the lead coder on the team. But they can also get tangled up. They can fix an error while creating another. They can sound confident while quietly moving the problem somewhere else. In software development, that can be managed through testing, review and version control, but you still need people with the expertise in the middle to spot the ‘probably’ right moments. In healthcare, the equivalent safeguards need to be even stronger.

This isn’t an argument against AI or tech. Quite the opposite. AI has the power to help us understand care more deeply. It can reveal hidden risk and unmet need through its ability to compute more in the same time. It can help us act earlier. It can help uncover the causes that sit beneath poor outcomes. It can also help build more resilient connected systems. AI can be used internally on the system and its infrastructure design, to learn from decades of healthtech development to create a foundation for the future from the patch work of connected, half connected, and disconnected systems we have today.

This is where major investment in healthtech needs to improve next.

Not only AI companies. Not only analytics platforms. Not only national data infrastructure. This applies directly to the next generation of EPR suppliers as well.

For years, much of the market has competed on functionality, modules, procurement promises and large scale implementation programmes. Those things still matter. But the next stage of healthtech will be won by the companies that make the foundations work in a way that feels almost invisible to the people using them.

The future is not one system trying to do everything. It is an ecosystem of healthtech suppliers working together around the patient, the clinician and the pathway of care.

That ecosystem needs strong foundations and plumbing.

The patient record should feel like a single record, even if it is assembled safely from multiple sources behind the scenes. The GP record, hospital history, imaging, pathology, medication history, care plan, discharge summary, community notes and relevant social care information should not feel like separate worlds to the people trying to make a safe decision.

Technically, they may sit in different systems. They may be governed by different organisations. They may have different data structures, access rules and clinical responsibilities. But that is the point. The complexity should sit behind the scenes. The experience for the clinician should be simple, safe and useful, a foundation built on the international standards.

This is the analogy people instinctively make with their phone.

They do not need to understand every security protocol, identity service, API call, cloud service, backup process or software update behind the scenes. They open the phone, authenticate once, and the ecosystem works. Apps talk to each other. Information moves. Updates happen. Security is present, but it does not make the whole experience unusable.

Healthcare is more complex and the risks are higher. It cannot be reduced to consumer technology. But the expectation is still reasonable. If the technology in our pocket can feel joined up, intuitive and continuously improving, clinicians are right to ask why the technology in hospitals so often feels fragmented, slow and difficult to use.

A “future EPR” should be more like the operating system of care than a closed box that everyone has to work around.

In this analogy, the EPR is closer to Android or iOS. It provides the secure foundation, identity layer, core record, workflow environment, permissions, resilience and user experience. Around it, an ecosystem of applications can innovate. Specialist tools, AI assistants, remote monitoring, scheduling, imaging, medicines optimisation, research, ambient documentation and patient facing services should be able to connect safely and intelligently.

But that only works if the foundation is strong.

The data needs to be accurate. It needs to be available. It needs to be secure cloud enabled. It needs to be protected by strong governance and security. It needs to be accessible to the right people at the right time, with the right safety features, audit trails and permissions. It needs to move across boundaries without becoming unsafe, duplicated, delayed or unreliable.

“One login to rule them all” sounds like a convenience issue. It is not. It is a clinical safety, productivity and workforce issue.

Behind that simple experience is complex technology: identity management, role based access, authentication, information governance, cyber security, audit, device management and permissions. But the user should not have to carry that complexity. The clinician should not need four logins to understand one patient.

The same applies to cloud infrastructure. A move from on-premise servers to cloud should not be something a patient notices. In many cases, a clinician should not notice it either. The screen should look the same. The ward round should feel the same. The clinic will run as normal.

But behind the scenes, cloud changes what is possible. It supports resilience, always on services, safer backups, faster recovery, better analytics, scalable AI, more frequent updates and more secure monitoring. It creates the conditions for improvement without requiring every organisation to rebuild everything from scratch.

This matters because the current model too often feels like patching. Patch the system. Patch the integration. Patch the workflow. Add a portal. Add a spreadsheet. Add a dashboard. Add another login. Provider IT teams become experts in the system they are paying for more so than the people who work for the company they’re buying it from.

The next generation of healthtech should move away from that model. Updates should not require months of planning, hundreds of staff, major disruption and a massive jump in improvement that’s made to feel risky. In other sectors, secure updates, rollback, testing, monitoring and incremental deployment are normal. Healthcare needs a version of that maturity that is appropriate to clinical risk.

That does not mean reckless change. It means safer change.

It means version control. It means rollback. It means testing in real workflows. It means monitoring after deployment. It means clinical safety cases. It means listening to users. It means the ability to improve continuously without turning every improvement into a major event.

This is also where the future and AI connects back to the basics.

AI will not reach its potential in healthcare if it sits on top of fragmented records, poor data quality, clunky access, weak interoperability and workflows that do not reflect how care is actually delivered.

A powerful AI tool looking at incomplete data may simply produce a confident answer from a partial picture. An ambient voice tool that only helps populate a difficult system may reduce typing but fail to improve care. A predictive model that cannot connect to the real pathway may identify risk without helping anyone act on it. A summarisation tool that cannot see the full record may make missing information look like certainty.

The model matters. But the plumbing around the model matters just as much.

If the foundation is weak, AI becomes another layer of complexity. Another dashboard. Another portal. Another thing to log into. Another partial view.

If the foundation is strong, AI becomes more useful, safer and more humane. It can help clinicians understand the patient, the pathway, the next action. It can support care teams across organisational boundaries. It can make the system easier to navigate, rather than simply making the system hungrier for data.

That is why the future of healthtech will be won behind the scenes.

The winners will be the companies that make the basics work beautifully: accurate data, cloud enabled infrastructure, safe access, strong identity, interoperability, resilient services, usable workflows, secure ecosystems, faster updates, reliable rollback and trust.

Not because those things are glamorous.

Because they are what make everything else possible.

For the past 20 years the goal has been to make healthcare more machine readable. The goal for the future is to make healthcare more understandable, more humane and more effective.

Healthtech should not ask healthcare to become more like software. It should make software more like the realities of healthcare.