The Silent Bottleneck and the Purple Squirrel:Why AI in Europe’s pharmaceutical industry stalls for want of people, not money
The money is there. McKinsey puts the annual value potential of generative AI for pharma and medtech at 60 to 110 billion dollars. In a survey of more than a hundred industry leaders, everyone had experimented with it and close to a third had begun to scale. Five percent had turned it into a lasting, measurable financial advantage. Even so, more than two-thirds plan to raise their investment significantly. Money is not what is holding things back.
THE BOTTLENECK IS PEOPLE, NOT MONEY
You might expect the bottleneck to be the data, or the technology itself. Both are genuine hurdles, but not of the same kind..
Data issues, on the other hand, such as availability, rights, interoperability, quality, and regulatory constraints, cannot be resolved with money alone, only by people who understand them. The binding bottleneck is therefore neither capital nor data nor technology. It is the rare ability to combine domain expertise, data, and AI in individuals who can also lead teams. Without it, every other resource sits idle.
The World Economic Forum reaches the same conclusion across sectors: 63 percent of employers name skills gaps as the biggest barrier to transformation, followed by their own organizational culture (46 percent). A shortage of capital comes only seventh, at 26 percent. In life-science laboratories the trend is sharper. According to the Pistoia Alliance, the share of respondents citing a lack of skills as a brake on AI adoption rose from 23 to 34 percent in a single year. Boston Consulting Group reduces the weight to a rule of thumb: 10 percent algorithms, 20 percent data and technology, 70 percent people and processes.
WHY IT STAYS SILENT
A bottleneck this large ought to be loud. It is not, because nobody measures it properly. Official statistics have no category for such people. A scientist with data-science skills is counted as a chemist or as a data specialist, never as the hybrid. Companies themselves look no closer: in the McKinsey survey, a mere six percent had systematically assessed which AI skills they lack.
What is not counted does not appear in any workforce plan. The shortage takes effect all the same, under other names: as delay, as a pilot that never reaches operations, as budget spent without anything having changed.
THE MISSING PROFILE IS ONE, NOT TWO
The obvious reflex is to hire both a biologist and a data scientist, and seat them side by side. In many settings that works. In the demanding ones it does not. A model that helps decide which target molecule to pursue, or which process deviation is critical, needs more than two skill sets in parallel. It needs them fused in a single act of judgment.
The data scientist without domain knowledge builds a clean model around the wrong question. The scientist without data skills asks the right question and cannot test the answer. What is missing is the person who looks at a model’s output, senses that it is biologically implausible, and knows why. Hire ten data scientists and you have ten data scientists. You do not have a translator. The profile is sometimes described as π-shaped: two deep columns, domain and data, joined by a crossbar that we read as the AI-supported judgment to connect them. This is not a polymath. It takes two disciplines, not ten, fused at depth. Headhunters have a name for such a candidate: the purple squirrel.
In research the profile already has a job title: the computational biologist who writes production-grade code, the cheminformatician, and the research engineer with real domain depth. The number of people who are at home in both worlds at this level is growing more slowly than demand. And that demand no longer comes from pharma alone. The profile is also missing in a second place, one the debate rarely visits.
THE UNDERRATED SECOND ARENA: PRODUCTION
The debate is almost entirely about research, however, the same gap runs through the factory floor, where it is named far less often. The technology has long been available: connected plants, sensors, digital batch records, AI-assisted process monitoring. ISPE’s seventh Pharma 4.0 survey shows that precisely where AI comes in, pilots are most numerous and scaled applications fewest.
Europe raises the stakes. The draft EU GMP Annex 22, developed by the EMA together with PIC/S, the international inspectorate network whose members include the FDA and the MHRA, explicitly leaves responsibility for the critical GMP decision with the company and its accountable persons. The EU AI Act is in force; in July 2026 legislators postponed its obligations for high-risk systems to December 2027 and August 2028 but did not drop them. The production profile required is therefore not quality plus engineering plus data, but all of these plus non-transferable accountabilities. Anyone who starts building it once the rules bite is starting too late.
FRONT-RUNNERS BUILD BEFORE THEY BUY
The companies that are ahead build first and buy selectively. Among the industry’s AI leaders, upskilling their own scientists is the most common source of talent, well ahead of any external one. The reason is plain: these people already have the part of the profile that is hardest to acquire later, namely domain depth.
What stands out is how close to the domain even external hiring stays: AI biotechs, other pharma companies, academia. Only 21 percent come from the tech sector. Quantitative finance and aerospace tend to appear on such lists, but no documented pattern shows their talent moving into pharma. These fields share the mathematics with pharma. The mathematics is the easy half.
The realistic answer is a portfolio: develop translators in-house, hire the rare purple squirrels selectively, use partnerships to scale. Where no single person combines both skill sets, build the team that does. Three walls remain that no mix can knock down: upskilling will not produce frontier research, hiring will not fix retention, and outsourcing will not transfer regulatory accountability. The model-builder can be rented. Responsibility for the GMP decision stays in-house. Finding these people is one problem, and how quickly it is solved sets the pace at which AI moves from pilot to operation. Keeping them is another.
WHY THEY LEAVE IS RARELY ABOUT PAY
Base salaries for AI specialists are now set by the tech market, and the industry pays them. Equity is where the distance remains wide. For the very top tier, the handful of researchers the labs bid millions for, no pharma company joins the auction.
For everyone else, specialist life-science recruiters say senior-level candidates rarely decline because of pay. Three other factors are more decisive, and none requires a tech giant’s budget: a hiring process that takes far longer than at AI-native firms; an AI strategy that fails to withstand scrutiny from experienced candidates; and limited upside, with little or no equity.
Pharma’s strongest argument is purpose. A model that speeds up a cancer trial beats one that optimizes ad clicks. Purpose alone keeps no one, though. It brings people through the door, but what keeps them there is whether they can work as AI practitioners and deliver results. Europe will not win a salary race against the US labs. It does, however, offer what matters most for retention: purpose, stability, and quality of life.
LEADERSHIP DECIDES WHO STAYS, AND WHO FOLLOWS
Much of that comes down to leadership. The Chief AI Officer has arrived in the pharma industry but is not yet standard. Even among the front-runners, a dedicated role of this kind is the exception. More
often, AI leadership sits inside R&D (30 percent) or in a hybrid of central governance and embedded specialists (35 percent).
The less comfortable issue is succession risk. It is often reduced to a question of age, but three forces are at work. There is the conditioning of a leadership generation that learned to treat technology as a function to delegate. There is blocked succession, as long tenures slow the rise of the next tier. And there is demography, which hits the scarcest profiles hardest: on calculations based on Cedefop’s forecasts, the EU will need to replace around a third of its ICT professionals by 2035, mostly because of retirement.
No comparable forecast exists for leadership, production or quality. Look at the age profile of your own plant and quality leadership and you will usually see quickly whether the same wave is coming.
SIX MOVES
01- Measure the bottleneck before managing it. Which dual-skill profiles are needed in research and production, which exist, and where are the fault lines?
02- A portfolio, not a single bet. Build, buy, borrow, with the three walls in view.
03- Fix what costs strategy, not budget. Process length, a credible AI roadmap, AI people embedded in research rather than in a staff function.
04- Play Europe’s real cards. Make deliberate use of purpose, stability and quality of life instead of losing on salary comparisons.
05- Build the production profile ahead of regulation. While Annex 22 is still a draft and the AI Act’s high-risk deadlines are still running.
06- Build the next leadership generation deliberately. It does not emerge from a new title on the organization chart. Find and hire the purple squirrel.
WHAT TALENT WORK CAN DO, AND WHAT IT CANNOT
It would be dishonest to close this paper with the promise that good appointments will solve the bottleneck. Reskilling has a ceiling, hiring a limit, borrowing a price. And no appointment, however astute, replaces a credible strategy.
Something else stands out. Almost every one of these six moves goes beyond filling a vacancy: measuring is diagnostics, fixing what can be fixed is employer branding, building the next generation is succession planning. The talent question thus moves from the individual vacancy to orchestration. And orchestration is decided at the top.
ABOUT THIS PAPER
This short version is based on a full-length white paper in which every load-bearing figure is traced to its primary source; time-sensitive facts were checked in August and September 2026. The section on production also draws on our advisors’ own specialist publications. Roles like these rarely have a predecessor to define them. The purple squirrels who can fill them are seldom on the open market and rarely answer a job posting. Finding them and winning them over for our clients is our work.
