Healthcare AI: Moving From Hype to Measurable Value

Healthcare AI is mostly sold to executives based on how impressive it is. Deals start with how advanced the model is, the buyer's advantage in the race, the model's generative capabilities, and the like. What the sales team fails to address is the most important question after the deal is signed: how does this model solve a real problem, and what is the return on it.

This question became the foundation of a Signal Room conversation with Parth Gargish, a builder of AI-first strategies. His perspective was straightforward. There are many AI organizations, and everyone is doing it. The important question is, are you solving a problem? This is also the stand we take at Hutchins Data Strategy Consultants. The difference between a deployment that justifies its cost and one that uses up the budget is almost always the scope of the work, not how complicated the model is.

The Test Is a Real Problem and a Real Return

Gargish is right; AI is costly to build, and showing a return is a challenge. However, this cost is a good thing. It maintains a level of focus that is lost with the hype of new technology, which states that a problem and a potential solution must be identified before work can be done.

He was wise to distinguish value from spectacle. A use case does not need to be particularly novel to be valid. It can even be as mundane as automating a function in a single department. This may not generate top-line revenue, but it can increase employee productivity, or even speed time to market, to serve a new product or customer. In his opinion, small wins are what really matter. Those organizations that only focus on the large, showy, headline-generating wins tend to spend the most money and learn the least.

Consider the pointless project that attempts to show a commitment to currency. The example he gave was of a project aimed at updating to AI, selling the tool, and producing nothing of value. The end result is a loss of cash for the company. A tool that is purchased, not integrated into a process, and never actually employed by the employees of the company, is essentially a cash loss.

Employ the Workforce, and Then Provide the Tools

The most counterintuitive part of the conversation, unexpectedly, came from a technology CEO arguing that technology is not where to start. Gargish's AI-first strategy hinges on what he calls PPT: People, Processes, and Tools. The order is not arbitrary. People come first.

As the reasoning goes, only a tool delivers value when a process exists around it, and people are using it. Sunk costs on the purchase of all the tools on the market and skipping the two other parts of the strategy reduce the company's resources to paperweights. So, the important first steps are to be sure the staff is committed to the new strategy, and understand the direction and the roadmap for the planned changes, before any technology is deployed.

He made it clear that transitioning must occur from the top down. Leadership not only decides the route, they also ensure the people are confident of their roles in the future. Without leadership, the people closest to the work, those who understand which workflow can be optimized the most, remain silent as they feel that coming forward will lead to the voluntary destruction of their jobs.

Fear Is Inconveniently Real, and Acknowledging It Is the First Step

Gargish does not dismiss legitimate fear. He acknowledges that a number of jobs will be impacted. He wants people to view his statements in the context of the bigger picture. This will not be the first time jobs will be impacted, nor will it be the last. The people who survived the previous restructures were the ones who were adaptable and learned new skills. It is his belief that leadership truly owns this issue. He believes leadership should reskill people and stand up the new roles before the old ones disappear, so what employees hear is a career path and not a layoff. Leaving the fear unaddressed is the real problem.

People overlook the importance of trust, and it resides among the mid-level managers, the ones running the teams the whole plan depends on. Gargish said that the CEO has to first bring in those mid-level managers as real stakeholders and not people who have been sent a memo. Real comfort among the managers will translate to real comfort among the employees. Only if managers are given a real stake will they truly take ownership of the task.

Purchase Help Where You Do Not Have the Know-How

Another thing he highlighted was the danger of trying to do everything yourself. He has a tech company, and he will be the first to say that he would not attempt to build a factory himself. He would call someone who had already done that. It is quicker than doing it yourself, and it avoids the expense of figuring out a steep learning curve. AI works in the same way. Bring in the expert who knows the field to set it in motion, then when you understand it, bring it in-house. It is probably cheaper than trying to figure it out and realizing you could have saved a lot of time if you had just called in the expert.

Constructing anew is not always a requirement. He used the case of the early mobile touchscreen phone to illustrate how a deliberate incremental improvement was the basis of a change that was not about constructing a solution from the ground up. Given the proliferation of open-source models, along with the myriad potential tools, the quickest route to value is likely to be the integration of existing models and tools in the service of a genuine problem, rather than the construction of a new solution from the ground up.

Make Sure a Human Is Deciding

While he was upbeat about tools, Gargish was quite specific about not allowing tools to make decisions. He was excited about the prospect of a general assistant tool always being accessible, highly personalized to him, and able to bring up a list of options. However, he is against making decisions with a tool. He argues it should be used to bring up options and display the possible courses of action, and keep the decision with the person. There is always a person involved.

With regard to healthcare, that line is not a preference, but a necessity. The utility of a model is evaluated on the quality of the decisions it enables, and the responsibility for those decisions cannot be shifted to a model. A company that considers AI to be a source of options rather than a source of decisions is also the one most able to justify and protect how a particular decision was made.

How Hutchins Approaches Healthcare AI Value

Our work starts where the marketing skips ahead — with the question of which problem a model is meant to solve and what it is expected to return. We help organizations scope use cases that are small enough to deliver and concrete enough to measure, build the process and the people-readiness around them so the tools are actually used, and keep human judgment at the center of any decision the model informs. Much of this is the same discipline we describe in healthcare AI consulting, where the failure pattern is almost always a program that started from the technology instead of the problem; it also depends on the AI literacy that lets staff use these tools well and on the responsible-AI guardrails that keep the decision accountable. These themes run throughout The Signal Room podcast, where practitioners describe what it takes to move past the hype and reach value you can name.

Authoritative sources

• WHO — Ethics & governance of AI for health
• NIST AI Risk Management Framework
• Agency for Healthcare Research and Quality
• HIMSS
• NEJM AI