AI in Healthcare & Pharma Summit: Why AI Models Don’t Scale

Every industry conference has its tone, but the AI in Healthcare & Pharma Summit in Boston had something different — a sense of collective impatience. Not the negative kind, though. The kind that comes when people know the technology is ready, and the bottlenecks are now human, cultural, and operational.

The event drew researchers, data leaders, clinicians, and pharma executives who weren’t there to admire AI’s potential, but to figure out why so much potential still doesn't make it into practice.

Over two days, speakers and attendees kept coming back to a handful of themes. Here are some of the big ones we heard repeatedly:

How to move AI out of innovation labs and into the enterprise value chain
Breaking down data silos with federated learning and privacy-preserving architectures
Fixing the messy reality of document and data hygiene before deploying GenAI
Using low-code/no-code tools to empower non-technical teams
Measuring ROI in a way that resonates with both clinicians and CFOs
Building the governance, ethics, and operating models needed for AI at scale

In this article, we’ll unpack some of those and explore what leading organizations are doing right now to make operational impact with AI. Let’s dive in.

Why AI Stalls and How to Break Through

You’ve probably heard about this before. And there’s a fair chance you’re going to hear this time and time again.

Most AI pilots that succeed technically go nowhere operationally. These initiatives frequently stall in what many call “Pilot Purgatory,” a limbo where promising models never mature into scalable ones. Why? Because they fail to mesh with the living reality of clinical and operational workflows.

And it’s nothing new – the same problem was stressed out at this year’s AI & Big Data Expo.

An organization’s AI maturity is defined not by its proof-of-concept wins, but by its command of Implementation Science. This means one thing: the biggest barriers to AI adoption aren’t technical; they’re human.

Continuous Implementation Science

Most change management frameworks were invented for linear rollouts (e.g., implementing a new Electronic Health Record system). But AI doesn’t behave linearly. It adapts. It evolves after deployment. It requires organizations to shift from episodic change to Continuous Implementation Science.

For mature organizations, readiness is no longer a one-time milestone. They are adopting new mechanisms to keep pace:

Dynamic User Training. Static manuals are out. AI-assisted coaching is in – training that adjusts to individual proficiency levels.
Agile Feedback Loops. Clinician or researcher feedback needs to move straight into model refinement and workflow adjustments. This prevents the familiar drift between algorithmic intent and real-world use.
Leadership Alignment. Forward-thinking executives focus on cultivating AI-ready cultures. Their impact is measured not by procurement budgets, but by adoption metrics and workforce capability.

Before we jump into the successful use cases presented at the conference, explore more AI in healthcare use cases in our blog.

Case Study: Moderna and the Architecture of AI Fluency

Few organizations embody cultural AI maturity like Moderna. Built on a digital-first foundation, Moderna invests not merely in tools, but in its people. Kevin Anderson, Sr. Director of Clinical Operations, describes AI Fluency as an operational capability distinct from digital literacy.

In a world where cloud providers and open-source models have commoditized access to technology, Moderna recognizes a different competitive advantage: a workforce trained to apply, evaluate, and govern AI responsibly.

A big milestone in Moderna’s AI maturity journey is mChat, a custom internal instance of ChatGPT built on OpenAI’s API. Crucially, this wasn’t exclusive to data scientists. It was rolled out enterprise-wide.

Adoption: Since early 2023, mChat has reached over 80% internal adoption.
Strategic Intent: The company’s aim is full generative AI proficiency across all employees. CEO Stéphane Bancel positions this shift on par with the arrival of the personal computer.
Operational Impact: By embedding role-specific GPTs in functions from legal to manufacturing, Moderna augments every role. This approach enables the company’s aggressive plan to launch 15 new products in 5 years without scaling headcounts.

To move beyond generic or repetitive use cases, Moderna created the Clinical AI Innovators Network (CAIIN). This multidisciplinary forum offers a protected environment where staff can apply AI in clinical contexts:

Structured Experimentation. CAIIN establishes guardrails for testing prototypes of Agentic Automation, such as AI agents that draft protocol amendments or monitor site compliance. Teams can experiment without jeopardizing trial integrity.
Role-Based AI Coaches. AI mentors guide employees through complex workflows, encoding institutional knowledge directly into the system. This dramatically reduces onboarding time and supports a hybrid human+AI operating model.

Anderson outlines a framework for developing enterprise-wide fluency:

Exposure – Broad access to tools like mChat encourages experimentation and demystifies the technology.
Experimentation – Programs like CAIIN legitimize trial-and-error learning.
Enablement – Organizations must provide the compute, data access, and support required to scale what works.
Ethics – Fluency requires governance. Employees learn to assess outputs for bias, hallucination, and regulatory adherence.

If your organization is navigating similar challenges or looking to accelerate its AI maturity, we’re here to help. Reach out today to discuss how to build AI fluency and scale real-world impact across your organization.

Case Study: Enterprise AI Acceleration at AstraZeneca

While Moderna often draws attention as a digital-native standout, AstraZeneca offers a counterexample — proof that even the largest, most established pharmaceutical organizations can build AI maturity at scale.

Shuja Mohammed, Head of Strategic Planning & Operations for AI & Data Science, described a transformation defined by the slow, systemic dismantling of silos and the embedding of AI into the company’s operational fabric.

By 2024, AstraZeneca recognized that sophisticated technology doesn’t guarantee sophisticated outcomes. In response, the company launched the Enterprise AI Acceleration Program to reskill and unify its global workforce around a shared AI capability.

The program introduces Bronze, Silver, and Gold AI literacy certifications. The approach is intentionally playful, but the impact is serious: by mid-2025, roughly 12,000 employees had completed at least one level of certification. Internal surveys show that 85–93% of staff report meaningful productivity improvements while using the new AI tools.

Biological Insights Knowledge Graph

If Moderna’s maturity is cultural, AstraZeneca’s is infrastructural. The company has invested more than $250 million into AI research and supporting architecture, culminating in one of its most ambitious undertakings: the Biological Insights Knowledge Graph (BIKG).

The BIKG aggregates billions of biological and clinical data points (genes, proteins, diseases, drugs) into a queryable network. The knowledge graph enables a move from hypothesis-led discovery (a scientist has a theory and tests it) to data-led discovery, where AI can surface relationships between disease pathways and molecular targets invisible to human intuition.

By integrating R&D, clinical, and commercial datasets, AstraZeneca builds a real-time feedback loop:

Commercial performance informs discovery priorities
Clinical outcomes refine target selection
Early research decisions incorporate downstream signals

If your organization is working to break down silos, build AI literacy, or operationalize enterprise-wide intelligence, we’d be glad to help. Reach out to explore how to accelerate your AI transformation — culturally, technically, and at scale.

Operationalized AI: When Maturity Gains Momentum

If organizational maturity is scaffolding, operationalization is the living structure that grows from it. AI is now expected to act as part of the organizational nervous system, influencing the full value chain, from early discovery to clinical execution to commercial strategy.

Modern AI Stack: Architecture That Scales with Science

In healthcare and pharma, the data landscape has long been defined by fragmentation. Each domain — research, clinical, commercial — built its own tools, formats, and data structures. Ittai Dayan, CEO of Rhino Federated Computing, argues that enterprise-scale AI requires a shift: infrastructure must scale with scientific complexity rather than struggle against it.

A mature AI stack unifies three historically separate domains:

Data Engineering – the movement, transformation, and preparation of raw data
MLOps – the lifecycle management of models, from versioning to monitoring
Governance – the regulatory and ethical framework that ensures responsible deployment

Bringing these pillars together is not easy, but Dayan suggests it’s the prerequisite for real-scale AI adoption.

Data Hygiene for GenAI Success

The GenAI boom has exposed a critical weakness in enterprise data practices. As Chris Huff, CEO of Adlib Software, puts it:

Most GenAI initiatives stall because the inputs aren’t clean, structured, or governed.

In pharma, high-value information is locked inside unstructured content —SOPs, batch records, clinical documentation, CMC files. Organizations with advanced AI strategies now treat document hygiene as foundational.

Modern enterprises deploy Validation-Rich Document Pipelines, which include:

OCR with accuracy thresholds and confidence gating
Full data lineage capture
Controlled, versioned documentation
Audit-ready data lakes
Traceability back to validated source records

This ensures that when a GenAI agent explains a manufacturing deviation or answers a regulatory query, every sentence can be traced to a specific document version.

Intelligent Data Automation: Closing the Gap with Low Code

Will Spendlove, VP of Product Strategy at Alteryx, said that the quickest wins in operationalization come from Intelligent Data Automation.

Across healthcare systems, data scientists still spend up to 80% of their time cleaning and prepping data. Low-code and no-code platforms allow nurses, pharmacists, analysts, and administrators to build workflows themselves.

Alteryx use cases include:

NHS Supply Chain: Automated complex tender evaluations, saving 1,000+ hours and shortening procurement timelines by two months.
BODi (Customer Health): Used Alteryx combined with Snowflake to build a “Customer 360” system, enabling sophisticated monthly promotions with fewer resources.

As you see, operational excellence emerges not from a single breakthrough, but from compounding efficiencies created by empowered teams. If you’re looking to modernize your AI stack or deploy automation that accelerates insight across your organization, we have the exact expertise you need.

Janea Systems provides end-to-end Data & Analytics Consulting services that help organizations operationalize AI with confidence. Our team supports you with:

Data strategy & architecture modernization
Data engineering & pipeline automation
ML lifecycle optimization & MLOps
Data governance frameworks
Enterprise analytics & dashboarding
Cloud-scale performance engineering

Get in touch with us to explore how Janea Systems can support your transformation (check the contact form below).

When AI Must Prove Its Worth

As the initial glow surrounding Generative AI begins to fade, AI must now justify itself in financial terms. Reports from Bain & Company and KLAS Research show that AI initiatives, once tolerated as exploratory “science experiments”, are expected to demonstrate measurable margin impact or lose funding.

Bicckie Solomon, Director of Pharmacy at HCA Florida North Florida Hospital, frames the core dilemma this way: healthcare AI must satisfy two masters — clinical improvement and financial return. This splits ROI into:

Hard ROI: cost savings, operational gains
Soft ROI: clinical outcomes, patient safety, care quality

The trick, Solomon suggests, is aligning both in a way that resonates with executives who expect AI to perform like any other capital investment.

AI rapid prototyping is a clever, tactical solution to this problem. By building a functional slice of the solution in 1-3 weeks, organizations can validate the business alignment before spending millions. This ensures that Hard ROI (time savings, revenue generation) before the expensive build phase.

Case Study: AI-Powered Pharmacy Operations at HCA Healthcare

For HCA Healthcare, AI ROI isn’t built on isolated point solutions. It emerges from shared infrastructure where many use cases sit atop a unified foundation, lowering integration costs and avoiding the “pilot tax” that sinks many AI initiatives.

HCA implemented a clinical surveillance platform across 177 hospitals, designed to flag sepsis risks, adverse drug events, and other high-acuity issues.

Hard ROI

The rollout yielded a striking annual ROI ratio of 1:12.9, driven by:

Reduced drug spend
Shorter length of stay
Fewer complications requiring costly interventions

Soft ROI

The average pharmacist response time fell from 13.9 hours to 2.6 hours, a shift with direct implications for patient safety.

Cultural ROI

Drawing inspiration from the U.S. Army’s 160th Special Operations Aviation Regiment, HCA embraced AI as a force multiplier, creating consistency and lowering variance in clinical decision-making.

Case Study: Commercial ROI at Novo Nordisk

AI-driven ROI isn’t confined to clinical environments. Shihan He, Machine Learning Engineer at Novo Nordisk, outlines a framework for AI-enabled commercialization aimed at closing the “post-approval gap”— the space between regulatory approval and adoption.

AI enables Novo Nordisk to:

Identify Bottlenecks. Machine learning surfaces barriers to patient access—prior authorization delays, payer behavior, geographic disparities.
Personalize HCP Engagement. Instead of static segmentation, AI predicts the next best action for each HCP based on behavior, timing, and context.
Enhance Decision Velocity. With platforms like Tellius, business teams can query data in natural language, reduce analytics bottlenecks, and accelerate decisions.
Drive Equity. Equitable access is not an afterthought—it is an optimization parameter. Models are explicitly designed to surface underserved populations and ensure therapies reach them.

Novo Nordisk’s approach shows how AI can deliver ROI beyond revenue by advancing access, reducing disparities, and aligning commercial activity with patient needs.

AI Reality Check: The Work Begins After the Pilot Ends

The RE-WORK AI in Healthcare & Pharma Summit offered a snapshot of where the industry really stands in 2025. The era of marveling at AI’s potential is over. What remains is the harder work of integrating that potential into culture, workflows, and data ecosystems that weren’t built with AI in mind.

To turn AI maturity into measurable value, organizations need more than models. They need the infrastructure, engineering discipline, and lifecycle management to support AI at enterprise scale.

Janea Systems helps build that foundation through AI & MLOps Services, including:

ML pipeline design, automation, and optimization
MLOps platform engineering for scalable model deployment
Cloud-native architecture for high-performance AI workloads
Model governance, monitoring, and lifecycle management
Integration of LLMs and GenAI into regulated enterprise workflows
Reliability engineering for mission-critical AI systems

If your organization is ready to accelerate its AI operations and build a durable competitive advantage, contact us via the form below.

If AI Is So Powerful, Why Is Healthcare Still So Manual?