Original post here.
Pharma’s investments in artificial intelligence are evident. The transformation is not.
Pharma does not have a technology problem. It has a transformation problem.
The billions being spent on AI are producing real activity: external partnerships announced, internal tools built, enterprise licences signed, certifications issued. What they are not reliably producing is scale. Pilots start, show early promise, then quietly stall. The big question is why.
The reasons are multilayered. Think of AI transformation as requiring a clear vision of what you are trying to become, clear strategic objectives for how you get there, and three enablers that should make it real: People, Processes, and Platforms. Pharma has invested heavily across all three, just not equally.
Vision Without Objectives
Most large pharma companies have something they would call an AI vision. What they rarely have is what should sit below it: specific, measurable objectives that would tell them which processes to change, how to prioritise investment, and what success will actually look like.
And the objectives that do exist are usually built for where AI is today, not where it will be in eighteen months to three years.
A clear strategy is necessary but not sufficient to stop pilots failing. The bigger issues live one level down, in the enablers
Imbalanced Enablers
People
Pharma has already clearly invested heavily in AI literacy. AstraZeneca has certified more than 17,000 employees through a tiered framework. Moderna's employees have built over 3,000 custom GPTs. Merck's GPTeal is accessible to 50,000 people globally. These are real programmes representing real investment. But AI literacy is the first rung of the people transformation ladder not the destination.
There is a meaningful difference between knowing how to use a tool and actually working differently because of it, and a larger difference still between that and redesigning what your role exists to do. Most pharma people AI maturity is sitting on rung one.
Sanofi is a useful example of an organisation moving at different speeds. I previously ranked them first overall on AI maturity, with some genuinely advanced work underway. Yet their people agenda is still described as 'familiarising our colleagues with AI' and 'instilling a culture of curiosity.' That “rung one” language does not mean Sanofi is immature overall. It means even the more advanced companies can be far ahead on platforms and use cases while still talking about people transformation in early-stage terms.
AI literacy is a prerequisite. It is not a strategy. The metric to watch is not how many employees complete an AI module. It is whether job descriptions, performance frameworks and org charts start to change.
Platforms
The platform deals are certainly impressive on paper. Billion-dollar co-innovation labs with NVIDIA. OpenAI, Anthropic and Microsoft Copilot agreements signed across major pharma companies. DGX SuperPODs. The infrastructure layer is genuinely maturing and helping deliver results, particularly in AI drug discovery.
But infrastructure alone is not transformation. The right platform deal can help deliver successful outcomes but the announcement is not the achievement. From tracking more than a thousand pharma, digital health and AI partnerships over several years, a consistent pattern emerges. The ones that progress beyond the first press release are usually the ones where someone in the business had already decided what would change before the platform arrived. The ones that stall are the ones where nobody had.
The Missing P
Process is often either absent from pharma's AI transformation or defined far too narrowly. This is the spiritual home of failure to scale.
A Roche data analytics leader made the point nicely: most organisations focus on a handful of AI use cases, but a typical workflow involves 20 to 50 tasks, and optimising one of them rarely delivers meaningful impact. The interdependencies between tasks mean that without restructuring the entire workflow, including asking which tasks no longer need to exist, you are adding efficiency at the margins and calling it transformation.
The dominant pattern in pharma is AI layered on top of existing processes rather than replacing them. Someone builds a tool to accelerate a task within a workflow nobody has redesigned. The task gets faster. The workflow remains the same. The net impact is negligible.
AI provides pharma with the opportunity to annoy its customers more often and faster
The commercial organisation is the clearest example. Many pharma companies are now using AI to produce marketing content faster. The content is more personalised, produced in higher volumes. There is one problem: most HCPs already find content from pharma commercial teams self-serving. AI provides pharma with the opportunity to annoy its customers more often and faster. That is not a content problem. It is a process problem. Nobody has asked whether the underlying model of content-driven HCP engagement should exist in its current form at all.
AI-assisted clinical study reports are now being produced in minutes rather than months, a real achievement. But even if you reduced that to femtoseconds, it would not meaningfully accelerate drug approvals without a more holistic view of the entire development process and AI's role within it. Faster drafting sitting inside an unchanged review and approval chain produces faster drafts, not faster drugs.
The reason is structural. The team that builds the AI tool is not responsible for changing how the business works. The team that runs the business process is not responsible for the tool. Someone is usually assigned to bridge the gap but rarely has the remit or influence to close it unless senior leaders are actively and continuously pushing for it.
Training delivers certificates. Platforms deliver press releases. Process redesign delivers friction, at least in the short term. The short-term rewards flow to the visible and the announceable. Process redesign is neither of those things.
What Good Looks Like
The companies beginning to pull ahead share one characteristic: they are redesigning workflows, not just deploying tools into them.
Genentech's lab-in-the-loop model is a nice example. AI does not analyse experiments, it guides them. The model predicts, scientists validate, the model improves. The workflow was rebuilt around AI's role rather than AI being inserted into an existing one. That is the distinction.
The question separating the leaders from everyone else is not "how can AI help with this process?" It is "should this process still exist?"
Pharma is not failing at AI. It is succeeding at the parts that are visible, measurable and announceable, the training programmes, the platform deals, the certifications. What it is largely failing at is the harder, slower, less glamorous work of asking which processes need to change, who owns that change, and what the business should actually look like when AI is genuinely embedded in how it operates.
That is the conversation many pharma companies have not yet had. The ones who start having it now will be the ones that matter in five years.
Gary Monk is an independent advisor focused on AI, digital health, and strategic partnerships in pharma and healthcare. He works with pharma companies on AI strategy, partnership intelligence, governance, and identifying scalable use cases across commercial and business functions. He also publishes widely followed analysis on pharma AI and digital health partnerships.
