Why Collaboration Is the Missing Ingredient in Healthcare GenAI

In August 2025, MIT Media Lab’s Project NANDA released a striking finding: 95% of investments in generative AI (GenAI) produced zero return. Faced with this data, healthcare leaders have a choice. They can interpret it as proof that GenAI is overhyped and retreat from innovation altogether, or they can recognize it as a warning and choose a different path forward.

The problem is not GenAI itself. The problem is how it is being pursued.

In their Harvard Business Review article, “Beware of the AI Experimentation Trap,” Nathan Furr and Andrew Shipilov echo the warning raised by MIT researchers, noting that today’s GenAI failures closely mirror the missteps of the digital transformation era a decade ago (Harvard Business Review. https://tinyurl.com/46sucdb7). Then, as now, organizations allowed thousands of disconnected ideas to bloom, hoping that one might become a unicorn. The authors argue instead for focused, well-funded initiatives grounded in real user needs and explicit pathways to return on investment (ROI). MIT’s research reinforces this conclusion, showing that the 5% of GenAI efforts that succeed follow a remarkably consistent playbook. They do not chase flashy demos or generic tools; rather, they build systems that learn, admit uncertainty, and integrate deeply into existing workflows. In healthcare— where complexity is the rule rather than the exception—this kind of disciplined focus is not optional; it is essential.

Now is the time to apply our subject matter expertise and ask better questions: What parts of our jobs create the most frustration, friction, and inefficiency? How can AI meaningfully reduce that burden? How can it better serve patients, clinicians, faculty, and staff—while also solving real business problems for the institution?

Answering these questions requires acknowledging a hard truth: AI is not magic. No single group—clinicians, administrators, technologists, or vendors— has all the answers. How GenAI will ultimately improve patient care and support healthcare teams can only be learned through iterative experimentation rooted in collaboration, with continuous input from patients, providers, administrators, and technical experts.

Yet too many efforts rely on off-the-shelf tools or flashy demonstrations that promise transformation but cannot scale. MIT’s findings show that leaders often pursue cosmetic applications—particularly in marketing—while avoiding the harder work of reimagining core clinical and operational workflows. Adding “AI” to broken processes does not create value; it simply makes inefficiency more expensive.

True ROI emerges when institutions are willing to lean into friction rather than avoid it. Research highlighted in Fortune shows that while 95% of GenAI pilots fail because they rely on generic, brittle tools, the 5% that succeed embed AI deeply into high-value workflows, build systems that learn from correction, and design pilots with scaling in mind (Fortune. https://tinyurl.com/3ak2eu7x). These efforts demand collaboration—across disciplines, across hierarchies, and often across organizations.

Eugene Woods, CEO of Advocate Health, describes AI as a strategic imperative that requires deep partnerships, not transactional vendor relationships (Harvard Business Review. https://tinyurl.com/2p9mfshf.) This reflects what Frans Johansson calls the Medici Effect: Breakthroughs occur at the intersection of diverse expertise. Advocate Health’s emphasis on co-creation, rapid-cycle decision-making teams, and direct access to senior leadership illustrates how collaboration turns experimentation into impact.

Healthcare does not need fewer AI ideas. It needs shared ownership, honest dialogue, and collective problem-solving. Collaboration is not a soft skill in GenAI— it is the infrastructure. Without it, GenAI will continue to underdeliver. With it, healthcare can finally move from pilots to progress, and from experimentation to real, durable value.