SPEAKER:
Stephen Wood, EMEA Regional Leader for Healthcare & Life Sciences, Microsoft
KEY TAKEAWAYS:
• 91% of pharma organizations cannot operationalize AI despite near-universal investment and intent
• Frontier firms report 88% growth strength versus 23% for laggards, a gap that compounds each quarter
• Pilot proliferation — not pilot shortage — is the primary scaling barrier in most organizations
• Governance infrastructure accelerates AI deployment; deferring it to move faster achieves the opposite
• AI transformation structurally requires CEO-level ownership; IT-led strategies produce laggard outcomes by design
"91% of organizations that we talk to say that they are struggling with scaling and operationalizing AI." Stephen Wood delivered that figure at Pharma 2026 without theatrical pause, because the number speaks for itself. The pharma industry has never allocated more capital to AI, never produced more pilot programs, never hired more data scientists. And it has never been less able to convert that investment into scaled commercial impact.
The paradox is measurable. With 94% of pharma organizations planning agentic AI deployments and McKinsey estimating that up to 85% of industry workflows are addressable by AI systems, the ambition-to-execution gap is not a rounding error. It is the defining strategic problem of this moment. The organizations that have closed it are pulling away fast: frontier firms are reporting growth strength at roughly 88% compared to 23% for laggards, according to Microsoft's cross-industry research. That is not a performance variation. It is a structural divergence, and it is widening. The question Wood spent his session answering is why, and the answer has nothing to do with which model a company has licensed.
Pilot Accumulation Is Not a Pipeline
"I worked with an organization, who very proudly told me that they have 422 use cases that they were examining. I was overwhelmed, and I thought, how on earth can you possibly find your way through that?"
That organization is not an outlier. It is the industry's modal condition. Pharma is structurally predisposed to use-case proliferation: decentralized commercial operations, siloed therapeutic franchises, and regulatory complexity that makes every function see its own distinct AI opportunity. The result is not a pipeline building toward scale. It is a graveyard of disconnected pilots, each generating localized learning that never compounds into organizational capability. Breadth of exploration becomes a substitute for depth of execution, and the CIO presents a slide showing 400 initiatives while the business waits for impact.
The misallocation runs deeper than prioritization failure. Most organizations are investing in the wrong variable entirely. Wood argued that the fundamental unit of competitive advantage is not the AI model but something far less portable: "Intelligence is what's embedded in your organization. It's the people, it's their network, it's the data they use every day, it's the insights, it's the processes that you use to operate your business and organization. Those things are the important things that we need to unlock."
That framing is uncomfortable for organizations that have spent the past two years in vendor evaluation cycles. Every competitor has access to the same foundation models. GPT-4, Gemini, Claude — these are commodities available on identical commercial terms to every pharma company on the planet. Organizational intelligence, by contrast, is what creates defensible, non-replicable advantage: the proprietary data assets, the embedded process knowledge, the institutional expertise of 20,000 employees. Organizations that treat model selection as the strategic decision are optimizing the least differentiated component of the equation.
Wood drew a sharp line between two distinct phases of enterprise AI. The first wave was defined by productivity metrics, "how many hours do we save," with "AI embedded in the workflow" representing an entirely different organizational proposition. Efficiency calculations treat AI as an external tool applied to existing work. Embedded intelligence redesigns the work itself. The organizations that have made this transition are not running more pilots. They have identified the workflows where redesign creates compounding commercial advantage and concentrated their investment there. The 422-use-case organization has not found those workflows. It has built a structure that makes finding them nearly impossible.
The Governance Inversion
The instinct most executive teams bring to AI scaling is intuitive and wrong: governance slows things down, so defer it. Move fast, build pilots, establish proof of value, then worry about compliance infrastructure. Wood's observation from Microsoft's engagement across the industry inverts this logic directly. "Building a trusted environment, a trusted platform, enables you to go faster. And that's what we're seeing."
This is a reported pattern, not an aspiration. Organizations that invested early in AI trust infrastructure, including agent identity management, observability frameworks, and security architecture, are deploying faster than organizations that deferred those investments. The mechanism is straightforward: every enterprise AI deployment eventually hits the governance wall. Organizations that built the wall into the foundation cleared it before deployment. Organizations that deferred it hit it mid-scale, at the worst possible moment, with the highest possible disruption cost.
Pharma should have a structural advantage here that it is almost entirely failing to exploit. The industry has operated within some of the world's most demanding regulatory frameworks for decades. GxP compliance, pharmacovigilance reporting, adverse event documentation — these are not burdens that pharma merely tolerates. They are competencies pharma has built into its organizational muscle memory. The institutional knowledge required to design AI governance architecture, including risk stratification, auditability, and change control, already exists inside most pharma organizations. Companies that recognize this can convert existing regulatory capability into AI scaling speed. Almost none are doing so deliberately.
"AI is definitely not an IT question or an IT solution. It has to be a cross — and that's the center of excellence type model that I was talking about." The organizational reflex to assign AI transformation to the technology function is not a neutral governance decision. It is a structural choice that predetermines outcome. IT organizations are built to manage systems, not redesign commercial workflows. Assigning them ownership of AI transformation is the organizational equivalent of asking the facilities team to lead a product launch: the function is real, the capability is genuine, and the mandate is wrong.
The frontier firms Wood has observed have made a different structural choice. In these organizations, "chief AI officers sit and report to the CEO directly," making AI transformation "a CEO-sponsored engagement and opportunity." The reporting line is not ceremonial. It signals cross-functional authority, resource allocation priority, and the organizational permission required to redesign workflows that cut across business units. Without that authority, AI transformation stalls at the boundary of every function that controls a critical data asset or process step. The org chart predicts the scaling outcome, and most pharma org charts have already made the decision.
Discover more on this topic at Pharma Commercial Data & Tech Europe 2026 (4-5 November, London) Europe’s collaborative home for data and tech pioneers. Visit the website here.
The Commercial Model Frontier Firms Are Building
McKinsey's estimate that "up to 85% of workflows within the pharmaceutical industry can be transformed, enhanced, or automated by agentic AI" is frequently cited as an opportunity statement. It is more accurately read as a competitive threat. If 85% of workflows are addressable and frontier firms are already redesigning them, organizations still running AI as a chatbot layer on top of existing CRM architecture are not behind on technology adoption. They are behind on commercial model design.
The evidence that this is already operational rather than theoretical is specific. "The Ontada example — actually 150 million oncology documents that were previously under-researched and underused, and now 70% of those are being used to surface insights." This is not a proof-of-concept. It is a production deployment at scale, converting a previously inaccessible data asset into active clinical and commercial intelligence. The competitive implication for organizations sitting on comparable proprietary data, including claims histories, patient registries, and real-world evidence repositories, is direct. Those assets are either being activated or they are becoming liabilities as competitors build intelligence platforms around equivalent data.
The maturity trajectory for commercial AI runs through three distinct patterns. Pattern one, AI that answers questions and surfaces information on demand, is already table stakes in frontier organizations and rapidly becoming baseline expectation across the industry. Pattern two, human-directed agents executing specific commercial tasks within defined parameters, is where most frontier firms currently operate. Pattern three is where the commercial model itself changes: autonomous multi-agent systems connecting market access, medical affairs, marketing, and field force workflows end-to-end, executing across functions without human direction at each step. Organizations still building pattern-one capabilities are not competing for pattern-three advantage. They are competing against where frontier firms were eighteen months ago.
Wood noted that roughly 41% of leaders already see AI as an opportunity for rationalizing resources, a figure indicating that workforce implications are being factored into commercial planning even as the full transformation remains early-stage. The honest read on that statistic is that workforce decisions are being made based on pattern-one and pattern-two assumptions about what AI will do. Pattern-three commercial architectures will require different talent profiles, different incentive structures, and different definitions of what a commercial role actually is. Organizations planning workforce strategy around today's AI capabilities are planning for a commercial model that frontier firms are already moving past.
"Models will come, and they will go, and there will be a new one next week. What doesn't change is making sure that we're thinking about those five components of how do we build an intelligence platform within our organization." The organizations positioned for pattern three are not the ones that selected the best model in 2024. They are the ones that built the organizational infrastructure, including data architecture, governance frameworks, cross-functional ownership, and embedded workflows, that makes any model deployable at scale.
The Structural Audit Most Organizations Are Avoiding
Wood's evidence, taken in aggregate, points to a conclusion that most pharma organizations are actively avoiding: the 91% scaling failure is not a temporary condition that resolves as technology matures. It is a structural state that persists until organizations make architectural changes most are currently unwilling, not unable, to make. Unwillingness implies a decision that can be reversed. Inability implies a capability gap that can be filled. Most pharma organizations have the capability to make the required changes. They lack the organizational will to absorb the disruption those changes require.
"The challenge that everybody is having is how to scale that impact. And that's not just a technology or a model answer. That's actually how do organizations reshape the way they work in order to be able to maximize the AI at scale?" The organizations answering this question correctly are the ones building frontier-firm performance. The organizations answering a different question, which model, which vendor, which pilot, are building the 91%.
Three structural indicators separate organizations that will scale from those that will not. The ownership test: does AI transformation report to the CEO or to the CIO? If the latter, the governance structure has already constrained the outcome, because cross-functional workflow redesign requires cross-functional authority that no IT organization possesses. The concentration test: can the organization name its three highest-impact AI workflow redesigns with confidence, or does it maintain a spreadsheet of hundreds of use cases? If the latter, it is in the pilot graveyard, and adding more use cases deepens the problem. The governance-speed test: has the organization invested in AI trust, identity, and observability infrastructure, or deferred it to preserve velocity? If deferred, it is structurally slower than competitors that invested, and will discover this at the worst possible moment.
Wood offered a deliberate note of calibration, comparing confident predictions about AI's workforce impact to early speculation about the Industrial Revolution's effect on farming. Precise forecasts remain premature. The organizational imperatives, however, are already visible.
The 88% versus 23% growth gap is not static. Every quarter a pharma organization remains in pilot-accumulation mode while frontier competitors embed AI into redesigned commercial workflows, the distance compounds. The technology required to close it is available to every organization in this industry on identical terms. The organizational architecture required to close it is available to no one by default. It has to be built, sponsored at the CEO level, and protected from the organizational gravity that pulls every transformation initiative back toward IT ownership and use-case proliferation. Pharma companies that treat that gravity as a technical problem to be solved with a better platform will remain in the 91%. The ones that treat it as a leadership problem, requiring a different kind of decision from a different level of the organization, are the only ones with a credible path out.
To get you highlights of Pharma 2026 faster, we are using generative AI technology to summarise the transcripts of the sessions. If you have any feedback about the summary, please contact lucy.fisher@thomsonreuters.com.
Discover more on this topic at Pharma Commercial Data & Tech Europe 2026 (4-5 November, London) Europe’s collaborative home for data and tech pioneers. Visit the website here.
