When I finish an operation, I debrief. I talk through what I found, what I did, what the outcome was. I dictate notes. I write summaries. I generate, in the space of ten minutes, a rich and detailed account of a clinical encounter that took me years of training to be able to produce. And then its utility fades, stored in a PDF, an EMR, a dictation system. It doesn’t become research. It doesn’t inform the next surgeon who operates on a child with the same condition. It doesn’t contribute to the aggregate evidence base that could, over thousands of procedures, tell us which approaches lead to better outcomes, and which don’t. It just… sits there. Isolated. Unstructured. Unusable. Rich data that never becomes evidence.
I’ve been thinking about this problem for a long time. And I keep coming back to the same uncomfortable comparison: retail figured this out decades ago.
What retail understood that healthcare hasn’t
Whether you walk into a physical store or browse an online one, retailers are capturing data constantly. What catches your attention. What you pick up, click on, compare, or abandon. How long you spend in a particular aisle or on a product page. What you ultimately buy, return, review, or come back for later.
This data drives everything: inventory decisions, pricing strategies, store layouts, staffing, website design, personalisation, product development, and marketing. The modern retail experience for both physical and digital shopping is built on a foundation of relentless, systematic data capture. It is not glamorous infrastructure, nor the headline innovation customers notice. But without it, none of the innovation built on top of it would be possible.
Retail didn’t solve the customer experience problem first and the data problem second. It solved the data problem first, and everything else followed.
Healthcare has done it almost entirely backwards. We’ve invested enormously in clinical decision support, AI diagnostics, genomic medicine, and precision therapeutics. These are genuine advances. But the foundational layer of capturing what actually happens during every patient encounter remains, in most Australian hospitals and clinics, essentially unsolved.
An estimated 80% of clinical data exists only in unstructured formats. The average specialist consultation generates observations, reasoning, decisions, and outcomes that are spoken, scribbled, or dictated and then largely stranded in formats that no analytical system can easily use. The registries that exist depend on coordinators manually re-entering data that a clinician already produced once. It’s the equivalent of a retailer having their staff manually transcribe every sales receipt by hand at the end of each day.
The patient journey we’re not capturing
Think about what a complete patient journey actually looks like for someone presenting with a complex condition. They describe their symptoms to a GP. They’re referred to a specialist who examines them and forms an initial diagnosis. Tests are ordered: blood work, imaging, pathology. A treatment decision is made, explained, and consented to. Surgery is performed and debriefed. Recovery is monitored. Follow-up appointments happen over months or years.
At every single one of these touchpoints, clinically rich data is generated. Most of it is captured in some form. Almost none of it is captured in a form that allows it to be aggregated, analysed, and used to improve the care of the next patient who follows the same path.
No mature retail business would tolerate this. In healthcare, we’ve largely normalised it.
The tools to change this now exist
I want to be direct about something: this is no longer primarily a technology problem. The AI services required to capture spoken summaries, interpret clinical documents, process forms, and ingest unstructured data into research-grade records exist now and are commercially available. They are not experimental. They are not five years away.
I know this because I am using a working prototype in my own theatre right now. After every complex procedure, I dictate my post-surgical summary. An AI pipeline transcribes it, extracts the relevant clinical entities, and automatically populates structured registry fields. No manual entry. No coordinator. No data lost. The whole process adds nothing to my workflow because it intercepts the workflow I already have.
The registries that exist today have made significant contributions, built by dedicated clinical teams who care deeply about patient outcomes. But the challenge facing every one of them remains the same: how costly it is to capture data through manual interfaces, how much potentially valuable data never gets captured at all, and how difficult it is to evolve a system that was built to be static. That’s not a reflection of the people running them. It’s a description of the systems available to them today.
What changes when we get this right
The retail analogy has limits, and I want to acknowledge them. Clinical data carries privacy obligations and consent requirements that don’t apply to a loyalty card. The consequences of getting it wrong are measured in patient privacy and trust, not customer churn. These are real constraints that deserve heightened respect.
But the privacy frameworks already exist. The consent models already exist. What’s been missing is the capture layer, the solutions that take the data clinicians already generate, and route it somewhere useful, securely and systematically, without adding to their workload.
A modern registry can offer so much more for our future healthcare system.
The AI services I’ve focused on here are about removing friction from data capture, intercepting workflows that already exist. But there’s a more significant horizon. The machine learning models that will power the next generation of clinical decision support, predictive diagnostics, and personalised treatment protocols are only as good as the data they’re trained on. Unstructured, incomplete, or context-stripped clinical data from EMRs, legacy registries and disparate data sources risks training our future AI models on an incomplete picture of clinical reality. It undermines the foundations of tomorrow’s healthcare AI. High-fidelity, structured, contextually rich clinical data, captured systematically across every encounter, is not just the raw material for better registries today. It is the evidence base on which the intelligent healthcare system of the future will be built.
When we get this right, and I believe we will, the compounding effect will be significant. Every clinical encounter becomes a data point. Every registry grows richer and more analytically powerful over time. The patterns that only emerge across thousands of patients become visible years earlier. The next surgeon operating on the next child benefits from the aggregate experience of every surgeon who operated before them.
That’s not a vision for the distant future. It’s a description of what becomes possible when healthcare finally decides to capture what it already knows.
Authors
Dr Sanjeev Khurana is a paediatric surgeon based in Adelaide, co-founder and a board member of Data Dissect, an Australian clinical quality registry platform operating under Neo Data Pty Ltd.
This article is co-authored with Denis Maguire, Co-Founder and COO, Neo Data Pty Ltd (Data Dissect).
Have your say with our poll question: Are unstructured clinical records limiting the future potential of healthcare AI?
Add Comment