How to rebuild the pharma operating model for an AI era

In the early 2000s, AstraZeneca’s gefitinib program in lung cancer was widely viewed as disappointing based on overall trial results. Yet buried within the data was a striking signal: a subgroup of patients experienced dramatic tumor responses – patients later shown to carry EGFR mutations. The finding became a landmark in biomarker-driven oncology, proving that the right therapy can look ineffective in the average patient while being transformative for the right subgroup.

Yet it took years to develop the companion diagnostics and reshape trial design around it. In the meantime, up to around 24,000 mutation-positive Americans likely missed access to the therapy.

For healthcare leaders, AI offers a way to detect these signals and point researchers toward the biomarkers worth investigating. And it can achieve in hours or days what once would take years.

We are now moving beyond the initial AI hype and into the complex reality of operationalizing it. While the technology is here, with pharma embracing AI across discovery, clinical operations, commercial, and manufacturing, what most companies still lack is an operating model that can act on what AI surfaces. Because the technology, it turns out, is the easy part. The hard yards come with operating model change.

We’ve seen first-hand what happens when the operating model doesn’t change in line with AI investment. At one biotech company, the data needed to refresh a new AI model sat behind compliance reviews with no framework for what was permissible to share. Medical and patient support teams didn’t trust where predictions came from. Governance was rooted in old assumptions. Compliance defaulted to ‘no’; risk approvals to ‘wait’. The model sat unused.

Move 1: Break down the walls between enterprise knowledge and decisions

Across most large pharma companies, geographic teams, therapeutic areas, functional teams, and platforms each carry their own data, their own systems, and often their own AI tools – all separate from other parts of the business. Another wall separates internal data from external real-world evidence, payer signals, and partner data. This exists for safety, regulatory, and governance reasons. But it also stops information, decisions, and signals from moving freely across the business.

The difficulties created by proprietary enterprise knowledge silos is not new, but AI changes the fix. An “AI superhighway” around the outside of the business - a connecting layer rooted in a robust enterprise knowledge architecture, with entries and exits into systems that need to be reached - lets AI agents and human decisionmakers reach what sits behind the walls.

With that highway in place, the picture changes. Clinical operations data, translational biomarker assays, real-world evidence from EHR partners, and regulatory precedent can now connect. AI flags a responder subgroup in days rather than in a decade, linking the trial signal to the biomarker, the patient population, and the regulatory path that gets the therapy approved.

Move 2: Replace linear, stage-gated decisions with parallel ones

With the connecting AI layer in place, the operating model itself can change. Pharma’s operating model has long been organized as a relay race. Discovery hands a candidate to development; development hands a drug to manufacturing; manufacturing hands a product to commercial. Each handoff carries a stage gate and a wait time measured in quarters.

AI changes the math. The FDA real-time clinical trials initiative is one signal: continuous monitoring of study execution from day one, instead of the quarterly review-and-amendment cycle. This means teams that used to wait two years for a readout can decide as data comes in, and trials start new arms to test fresh hypotheses while existing arms run.

It also enables cross-functional teams to work in tandem rather than in serial handoff. Centralized protocol authoring tools let commercial, regulatory, and clinical operate on the same evolving draft. This is the operating-model shift: a team working concurrently on discovery, development, manufacturing, and commercial shaping decisions and responding to live data together, in real time, rather than waiting for handoff, confident in the provenance and reliability of the knowledge they are acting upon.

Move 3: Augment human ingenuity across teams

The third move is about people. In a pharma company, scientists, engineers, and data managers inside the line of business should be working alongside technology partners with the technical expertise to translate business problems into AI solutions. AI doesn’t just split the work; it supercharges scientists and SMEs, allowing human ingenuity to focus on complex judgments rather than data processing.

The SMEs understand the workflow and the regulatory tolerances, and have a clearer view of which judgments AI should take on and which must stay with humans. This move centers on cultivating AI leadership inside the business, rather than parachuting in central AI experts. Done well, this means embedding AI experts alongside operational leaders, with clear accountability and close connection to the work. The AI, and the experts who can support its proper use, are part of the team.

AI also changes how teams work together. In dynamic demand planning, for example, manufacturing has to trust the demand signals coming from Commercial and Market Access, and act on them even when that means a fast ramp up or down. Decisions that used to live inside one function now require newly connected teams to trust each other’s data provenance, insights, and judgements.

Without that trust, the operating model redesign does not hold. None of this works without changing how decisions get made. Equipping line-of-business champions only goes so far if the default answer to anything new is ‘no’, ‘wait’, or ‘commission a review’. What leaders need is clarity on what an AI model can and can’t do, where its outputs are sound, and where they need challenge. The leaders pulling ahead are building the infrastructure and governance that lets line-of-business teams experiment safely, learn fast, and spread what works.

What leadership needs to do now

Capturing AI’s value requires building a knowledge-driven AI culture.

First, start building the connecting layer. Pick the two functions whose work would most benefit from seeing across each other’s walls. Agree on the definitions and skills underneath. Build the entries and exits. Wins should land inside 90 days, with real models changing real decisions.

Second, move from periodic readouts to continuous monitoring on one major decision. Pick a question your team is currently waiting on quarterly data to answer, and start tracking the signals as they come in.

Third, name your line-of-business champions and equip the leaders above them. The best AI use cases are demand-led; they almost always come from the operational unit leader who has been shouting about a problem for years. Give those leaders the governance and the AI confidence to act on what their teams surface, fostering a culture of continuous learning and safe experimentation.

Pharma companies that get this right will design an operating model that makes better decisions, owned by the people closest to the work. The technology, in the end, was always the easy part.

The full article was first published in PharmExec.