Last year I engineered an advanced agentic AI pharma strategy development application.

One year later, after many illuminating discussions, the complexity of introducing such an offering became clear, and I decided that to shed more light I would produce a formal white paper based on a series of structured qualitative interviews with senior commercial leaders.

Here I set out the main findings with anonymised illustrative quotes drawn from the interviews and triangulate them with other published work by McKinsey, Bain, IBM, the World Economic Forum and others.

The successful adoption of AI in pharma is inextricably linked to the vision and culture set by leadership.

This is unsurprising, but the research showed that different leadership approaches generate very different internal environments: cautious, incremental and slow; or ‘trial-and-error’ and ‘fail fast’ approaches. Some tolerated or even encouraged failure.

“The CEO... pushed for us to really engage in leading [AI adoption]... and it changed the culture a lot.”

Crucial is the messaging that AI augments jobs, not eliminates them.

“AI is not going to take your job. The person who’s going to take your job is somebody who knows how to use AI.”

Leadership styles varied from urgent, CEO-driven competitive pushes to deeply grounded CIO-led initiatives. In some cases, the desire to move fast or be first appeared to overwhelm.

“There was no clear road map, but a lot of buzz from senior people”.

“There is a bit of FOMO”.

External research confirms that CEO-led, enterprise-wide AI transformations with phased maturity models deliver significant financial gains but rely on consistent executive advocacy.

AI adoption is overwhelmingly a people challenge, true of all other historic technology implementations, but the interviews provided a stark reminder. Leadership’s tone setting is vital, and where direction was hazy, workforce concerns intensified. Disparate AI literacy levels and uncertainty about AI outputs contributed to wariness among employees, making transparent communication and targeted training crucial.

“[In the future …] If you are not gold level in AI training, you’re out of the job.”

Organisations spoken to had divergent adoption styles: from mandatory certification programmes to looser approaches that encouraged individual-led experimentation. In many cases AI usage seemed to happen in user bubbles rather than as a coordinated change to workflow fundamentals, although where that change does happen it can be swift and dramatic.

“We have removed our first-line sales managers in Europe and US ‘like many other companies’. AI coaching has taken their place.”

There was little evidence of formalised change management, but in some cases there was clear encouragement of discussion and knowledge sharing to build confidence in AI’s contribution as part of a new culture.

“The speed at which you can learn and adapt is becoming more important than how good you are.”

By positioning AI as a complementary partner, delivering tailored training,
managing change transparently and embracing open communication, companies empower employees to harness AI confidently, thereby driving sustainable innovation.

Trust in AI insights grows with transparent validation and open dialogues. Industry voices emphasise hybrid talent models – generalists managing complex AI-human workflows alongside specialist domain experts – along with emergent roles such as AI coaches and autonomy auditors, pointing to systemic talent pipelines.

Respondents described their AI usage more as augmentation of existing commercial workflows – content creation, sales planning, administration – than one that overhauled operations. The respondents who described successful adoption emphasised a blended approach, reducing technical barriers via natural language interfaces and embedding human oversight (‘human in the loop’) to mitigate AI risks.

“[It’s about] the combination of human plus AI... not AI alone”
There were examples of radical workflow change, and one respondent described a process of ‘rewiring marketing’, although this was a work in progress and the precise meaning of the term remained unclear to the respondent.

One concern was that AI efficiencies are reshaping workforce structures, often reducing entry-level roles vital for career progression, demanding fresh talent strategies blending AI skills with experiential learning.

Governance is a critical component of AI adoption, and the research revealed mechanisms across a broad spectrum; from ponderous and risk averse:

“[Our] governance infrastructure is a dense and complicated process involving several committees, mainly due to concerns about data security and privacy” to dynamic and nimble, turning proposals round in days: “There was a digital team and we kept the decisions in there – to make certain that technically we weren’t doing things that [were risky].”

Legal and compliance teams play a vital role but can represent a barrier to AI experimentation. In addition, metadata gaps lowered AI content relevance, increasing mistrust and requiring considerable data sanitisation.

Respondents also outlined examples of ‘shadow’ AI use, where colleagues had been seen to be using AI applications that had not been regulated internally. This was clearly unacceptable to them, but there was a view that the lines were somewhat blurred, which was perhaps why nobody spoke out.

Progressive governance calls for layered controls – human in the loop for critical tasks, human on the loop for routine processes – with continuous oversight via cross-functional bodies integrating legal, medical, IT and compliance expertise.

Operational leaders must actively participate. This is not a technical issue to be left solely to the IT department.

“Leaving digital teams to execute was frustration... we had the tail wagging the dog.”