Healthcare runs on data. Every diagnosis, treatment, and referral leaves a record, and the assumption is that years of records make an organization ready for AI.
They usually do not. In this episode of AI ThoughtMakers, host Prem speaks with Rakshit, Product Manager at GeekyAnts, about what actually blocks AI in healthcare — the state of the data, the trust it carries, and the decisions a product manager has to make before any model is worth building.
Watch the full episode here:
The Problem Sits Before the Model
Intelligence is not the first problem in healthcare. Data is.
Records flow in from different hospitals, labs, and systems, which means different formats and different quality standards arriving in the same place. Many healthcare systems are still working on making that data consistent, and until it is, the data cannot be trusted. Adopting intelligence on top of it is the next level of problem.
Organizations sitting on years of history often read that history as readiness. The discrepancies are still there. Feed data with discrepancies into a model and it returns wrong predictions and wrong outcomes, which is a worse position than not having the model at all.
Why Trust Decides Whether Anyone Uses It
Consider a doctor working with two copies of the same information: the digital record in the system and the physical copy on paper. The first thing any user does is compare the values. If they match, trust builds. If they do not, the doctor cannot make a decision from the digital copy, and the system has lost its purpose.
That is why healthcare product decisions are about trust rather than usability. Every feature influences it, and adoption falls the moment it slips. A system that questions its own data gets abandoned even when it starts producing correct predictions later.
This is also where product management earns its place. The role stands between technical possibility and operational reality. Hospitals ask for dashboards, reports, predictions, and a long list of features, and someone has to decide which of them delivers value at that point in time.
What Organizations Actually Want When They Ask for AI Consulting
Most people assume AI consulting means recommending AI tools. Finding tools is the easy part- go to the market today and there are plenty of LLM models to pick from, and anyone can suggest one.
The work is in the decision. If a workflow is running and not producing the expected result, that is not automatic grounds for an AI solution. The first move is to analyze it and see whether modifying or redesigning the existing workflow gets you the result. If it does, the problem is solved.
AI consulting is as much about deciding where AI should not enter as where it should. AI does not create magic on contact, and it breaks unless the inputs going into the model are right.
Not every problem deserves an AI solution. Some deserve clarity.
What Failing Data Costs Here
In finance, bad data delays money. Production slips, and there is a financial loss at the end of it.
Healthcare differs at that point. Incorrect data reaches someone who has to make a decision with it, and that decision sits close to a human life. Once the people using the system start questioning the data, they stop trusting the system, and a system nobody trusts does not get used.
Organizations carry responsibility for this. Capturing data correctly and using it in the right way is part of the job, and an organization that is not confident in what it captures should not be running predictive analysis on it. Half-baked or incomplete data should not travel to a place where someone can make the wrong decision from it.
What the Data Layer Looks Like in Practice
On a health information exchange platform, the work runs like this. Data arrives from various hospitals, and the main job is checking its quality. Missing field segments and other inconsistencies get identified and reported back to the hospitals, who correct what they can and adjust how they capture records going forward.
The specifics come from how US healthcare systems generate files when a patient enters — ADT, CCD, and others. Fields and segments get missed often, and each miss is a data inconsistency downstream. Flagging them lets hospitals cover those gaps for upcoming patient records.
Once corrected data starts coming through, the FHIR system and the others around it can generate reports, run predictive analysis, and trigger alert mechanisms. Healthcare has to pass through that transformation layer first. AI adoption comes after it.
Where the Next Five Years Go
The transformation has already started, and most of the industry is still catching up on what happened rather than predicting what comes next.
Until now, the work has been understanding data after an event occurred. With clean, structured data, predictive analysis reads the patterns and tells you what is coming, which is where healthcare is heading and what teams should be ready for. AI starts creating value in healthcare when the data is adequate.
The Rapid-Fire Round
The episode closes with a rapid-fire round.
One thing hospitals underestimate? Data quality
Biggest myth about AI in healthcare? That AI can fix anything
One thing product leaders must learn? Clinical context
Does AI replace doctors? Never, it helps them decide faster
One word for the future? Predictive
What this Means for your Healthcare Product
The order matters more than the ambition. Get the data consistent, prove it against the source the clinician already trusts, and decide honestly which problems need a model and which need a better workflow. Predictive analysis is worth building toward once that groundwork holds.
Skipping to the model produces confident outputs from inconsistent inputs, and in healthcare that reaches a decision made close to a patient.
Our healthcare app development teams work on that groundwork — data quality, interoperability, EHR and EMR systems, and compliance — before anything predictive goes on top. If you are weighing an AI initiative against the state of your data, talk to us.
