Caution and Pharma’s AI Adoption: Understanding the Reasons Behind Uneven Uptake

Recently a colleague shared an anecdote that begins to explain the rationale underlying pharma commercial executives' somewhat uneven uptake of artificial intelligence (AI). This colleague, who had served on his company's AI governance committee, said the group's first instinct was to catalog every application already using AI for review.

It did not take long, however, to realize that nearly every function or initiative already touched AI, and reviewing it all at once became its own kind of paralysis. The committee shifted instead to writing guardrails for new use cases as they emerged.

This account illustrates one large reason why, during the past 2 years, enterprise-level integration of AI among pharma commercial teams has not followed the hockey stick–like growth curve that AI piloting has. Indeed, that instinct to slow down and ask, "What could go wrong?" is exactly the kind of caution these commercial teams need more of.

That brings us to a study that Brett Ramos, senior director of commercial operations and analytics for Neurelis, and I conducted among 100 commercial leaders across finance, information technology, sales, marketing, and analytics at biopharma companies.¹ We presented the results at the Pharmaceutical Management Science Association meeting this spring.

Who hesitates and why

We wanted to understand what was slowing AI adoption in an industry in which the stakes of a Health Insurance Portability and Accountability Act (HIPAA) misstep or a promotional compliance issue are higher than in most. AI adoption is moving quickly enough that some of the figures below have likely shifted since spring.

What has not moved as much is the underlying pattern of who adopts, who hesitates, and why. Several myths about AI adoption did not hold up particularly well against the survey results.

One of the most persistent was that enthusiasm predicts usage. Many respondents said they actively look for ways AI could improve their work, but a much smaller share had built it into their day-to-day routine. A subscription or a positive attitude was not enough.

Another myth was that department or age could predict who adopts and who does not. Some people in analytics had barely experimented with AI, whereas field representatives were asking for database access to build their own agents.

The same pattern showed up across age groups. In many cases, the deciding factor seemed to be whether someone had found a use case that saved meaningful time.

The results do not argue against investing in AI. They suggest that the rollout model matters.

A widely cited Massachusetts Institute of Technology report found that 95% of enterprise generative AI pilots deliver no measurable profit-and-loss impact.²

However, in a Snowflake study among early adopters, 92% reported positive returns, with life sciences companies earning back $1.43 for every dollar invested.³ The harder problem is organizational trust: A tool can work technically and still fail if employees do not know when to use it, what data they can bring into it, or who is accountable for the result.

Different starting points, same team

Our survey data point in the same direction: Adoption tracked with mindset more than role, which is a major reason a single company-wide AI rollout rarely works. The one-third of respondents who had built AI into their work sat alongside another one-third who were only experimenting, with the rest waiting for proof or using AI only when required.

That self-assessment may be generous, because people tend to describe their adoption better than their habits show. Still, three different starting points sat inside the same team.

To make sense of those differences, we grouped the responses into five personas out of eight identified by the survey. It is the same segmentation logic commercial teams already use to understand health care providers, just turned inward on their own colleagues.

The Governor, the one we heard about most in pharma, was focused on compliance, legal exposure, and what happens when a tool touches HIPAA-covered data. Whether the AI works mattered less to Governors than whether anyone could explain or defend the decision it helped make. The Enthusiast trusted the technology more than it had earned.

The Anxious Skeptic worried about what AI meant for their job. The ROI-er wanted a quantified pilot before broader investment. Finally, the Overwhelmed saw the potential but had no time to learn another tool.

These are not fixed identities. Someone may be a Governor when a project involves patient data and an Enthusiast when using AI for an internal summary. That is why rollout plans need to account for context, not just personality.

Mind the Governor

Unsurprisingly, the top barrier respondents named was compliance and regulatory uncertainty. That makes the Governor especially important.

Compliance concerns should be part of the design process before a tool goes live, rather than a review that happens after the fact. When compliance is built in early, caution can help define a workable use case instead of stopping the conversation altogether.

The survey does not tell us that concern about compliance is always the right instinct. It does suggest that treating that concern as a nuisance is unlikely to improve adoption.

The other top barriers, in order, were data quality, insufficient training, and a lack of approved use cases. Those barriers are as much about trust and process as they are about technology.

This is where the governance committee example from the opening becomes relevant. The committee did not need to approve every possible use of AI at once. It needed a process for evaluating the next use case, setting the right guardrails, and learning from what happened. For many pharma organizations, that may be a more practical definition of progress than a company-wide rollout.

References

1. The 2026 biopharma commercialization research report. Beghou. Accessed October 27, 2026. https://beghou.com/commercialization-that-works
2. Gultekin B. The radical ROI of gen AI. Snowflake. April 14, 2025. Accessed October 27, 2026. https://www.snowflake.com/en/blog/gen-ai-early-adopters-report/
3. Challapally A, Pease C, Raskar R, et al. The GenAI Divide: State of AI in Business 2025. MIT NANDA; July 2025. Accessed October 27, 2026. https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf

About the Author

Nicole Ventrone is a partner at Beghou.