Blog

From Hype to Impact: How Pharma is Actually Implementing AI and Unlocking the Value of Clinical Data

Written by Yulie Klerman | Jun 4, 2025 12:45:00 PM

Across dozens of conversations we had with pharma and biotech leaders, spanning R&D, medical affairs, clinical development, market access, digital health, and real-world evidence, one thing has become clear: AI is no longer a curiosity or side project. It’s a business mandate. 

Whether during executive roundtables, 1:1 meetings, or informal catch-ups at recent conferences, a consistent pattern emerged: teams are being pushed to move from experimentation to execution. The pressure is on to prove that AI can deliver measurable value at scale and in the real world. 

Here's what we’re hearing most often—and where the smartest teams are focusing their energy.

From Curiosity to Business Mandate: AI is Now Being Judged on Performance

AI is no longer in the sandbox. In every conversation, from digital to RWE to clinical teams, leaders are now accountable for ROI. The exploratory phase is over. What matters now is: Will this scale? Will it deliver value? Will it stick? 

The WSJ article covering Johnson & Johnson’s AI pivot (“J&J Pivots its AI Strategy”) triggered broad industry reflection. After piloting nearly 900 use cases, J&J retired 90% - not because the technology failed, but because only 10–15% of use cases delivered 85% of the value. Those that stayed focused on embedded, outcome-driven tools: internal knowledge assistants, AI-enhanced sales coaching, and selected discovery tools. 

That example mirrors what we’re hearing across the industry. The question is no longer what AI can do, but how to scale and implement it meaningfully. From R&D to medical affairs, teams are being asked: Does it work in the real world? Will people or more frequently healthcare providers use it? And can it show measurable business or clinical impact? 

The most grounded teams are now tracking three success metrics: first, can the use case be successfully deployed in real workflows; second, will it see adoption beyond a pilot; and third, does it deliver ROI or clinical value. 

AI is no longer something to pilot for its own sake. It’s about outcomes, integration, and impact. The bar has been raised from hype to performance. 

This is especially true for clinical AI use cases. Whether it’s surfacing missed diagnoses in rare disease, improving patient journey understanding in oncology, or identifying care gaps in chronic conditions, pharma teams want partners who can deliver outcomes, not just algorithms. 

Everyone’s Building a Data Strategy Team—Because the Ecosystem Is Overloaded

One striking trend: large pharma companies are investing in "Data Strategy" or "RWD Strategy" teams. These teams don’t just aggregate data sources, they're tasked with making sense of an increasingly crowded and chaotic vendor landscape. 

Many describe it as data chaos: the explosion of structured, semi-structured, and unstructured sources, from EHRs and labs to registries, PROs, claims, and social data, has made it harder, not easier, to navigate. What once promised insight now risks duplication and waste without clear alignment to business needs. The challenge isn’t lack of data; it’s knowing what data to use for which question. 

We hear this especially from medical affairs and HEOR teams trying to define real-world comparators, understand population characteristics, or identify treatment patterns. Without a clear framework, they’re left guessing what sources are fit-for-purpose. 

Teams that get this right are building dynamic maps of: 

  • What data supports which use case (e.g., label expansion vs. early detection vs. payer evidence) 
  • What gaps still remain in coverage (e.g., disease progression, functional status, behavioral data) 
  • What sources can be harmonized and scaled 

That’s a big mindset shift: from asking "Do we have data?" to "Do we have the right data, for this question, in this population, at this point in time?" 

From Structured to Narrative: Where the Real Insight Lives

When pharma teams say they "need data," they mean it in service of a goal. No one wants data for data’s sake. They want to: 

  • Understand patient journeys 
  • Identify care gaps 
  • Find undiagnosed patients 
  • Discover predictive factors for treatment response 

The problem? Much of what matters is missing from claims or structured EHR data. Patient populations have grown more nuanced. Treatments have grown more specific. And the information that differentiates one patient from another often lives in free-text clinical notes. 

To truly understand real-world patient populations, you need to read the story between the lines. For example: 

  • In our ASCO 2023 study, we showed how analyzing 150,000 oncology notes revealed where and why patients were dropping out of follow-up. This wouldn’t be visible in claims. 
  • In another project, we identified a 3–5x larger undiagnosed population for a rare disease compared to what structured data suggested, by detecting patterns across radiology, neurology, and progress notes. 
  • Our recent work published in Nature illustrated how analyzing unstructured NICU data helped reduce costs while improving infant outcomes—showing not just model accuracy, but clinical and operational value. 

We often explain it this way: structured data gives you the shadow. Narrative data gives you the whole figure – rich with context, nuance, and meaning. 

This depth of insight isn’t just academic – it allows teams to target patients more precisely, explain payer decisions more clearly, and surface unmet needs more reliably. 

From Patient Lists to Predictive Patterns: Use Cases Are Growing Up

Even a year or two ago, the focus was on surfacing the right patients. Today, the conversation is evolving: 

  • Can we detect clinical thresholds when patients become eligible for trials? 
  • Can we identify subtle signs of disease progression before it’s formally coded? 
  • Can we predict which patients are most likely to respond to treatment—and why? 

We’re supporting clients in: 

  • Identifying early signals of small-cell lung cancer months before diagnosis 
  • Uncovering why patients who meet label criteria are not being prescribed treatment 
  • Exploring predictive phenotypes by reverse-engineering responder profiles—i.e., identifying shared traits among patients who benefited from treatment, then using those traits to inform future targeting 

These are not abstract problems. They are real barriers to effective drug development, equitable access, and market performance. 

The Bottom Line

The AI conversation in life sciences has matured. We’ve moved from proof-of-concept curiosity to enterprise-level implementation. But the winners in this next phase won’t be those with the flashiest technology. They’ll be the ones who: 

  • Pair the right use case with the right data 
  • Connect clinical insight to operational value 
  • Turn narrative noise into structured understanding 

If your team is navigating these same challenges—whether around data strategy, unstructured data, or scaling clinical AI—let’s talk. The stakes are high, the tools are ready, and we know what it takes to go from proof-of-concept to practice.