Pharma's AI Skills Are Strongest Where the Stakes Are Lowest

Nearly 9,400 verified employees across 105 capabilities reveal a 14-point spread between using AI and using it responsibly.

‍

Pharmaceutical companies have moved faster on AI than almost any regulated sector. Very few can prove it is working.

Deloitte's 2026 Life Sciences Outlook found that only 22% of organizations report successfully scaling AI, and just 9% report meaningful financial returns. The usual explanations point at data infrastructure or model governance. There is a simpler one hiding underneath: most organizations have no objective view of whether their people can use these systems safely and well.

Across Data and AI readiness programs at leading pharmaceutical companies, nearly 9,400 of more than 12,000 invited employees verified their skills. Nearly 90 programs, spanning 105 distinct capabilities, ran in departments from commercial through R&D. The resulting picture is specific enough to act on, and one finding stands out above the rest.

The spread runs the wrong direction

Readiness reached 82% in foundational Generative AI skills. It fell to 68% in Responsible AI use.

Fourteen points separate what people can do with these systems from whether they know when not to. In most industries, that ordering is inconvenient. In pharma, it is inverted against the risk.

A misapplied model in a marketing function costs a quarter. The same misjudgment in a trial protocol, a safety signal review, a regulatory submission, or a patient-facing decision costs something the industry does not get to write off. The capability that carries the most consequence is the one the workforce is least prepared for.

"In pharma, the question isn't whether people are using AI. It's whether they can be trusted to use it well," said Kian Katanforoosh, founder and CEO of Workera. "Our data across leading pharmaceutical companies shows workforces are strongest exactly where AI is most foundational and weakest where it demands judgment, like Responsible AI. That's the gap that becomes a governance problem if you can't see it."

The sector is already asking the right question

Here is what makes the finding useful rather than damning. The same three capabilities surfaced at the top across every one of these programs: Generative AI Essentials, AI and Data Communication, and Responsible AI Essentials.

Responsible AI is not an afterthought in what these companies chose to measure. It is a priority they named themselves. The gap is not one of intent. It is the distance between deciding something matters and confirming your workforce has actually gotten there, which is exactly the distance verification is built to close.

Self-assessment cannot do this work. An employee confident in their judgment about AI is reporting confidence, not judgment. The two diverge most sharply in the capabilities that depend on knowing the limits of a tool rather than operating it.

Visible gaps close quickly

The other half of the data is more encouraging than the first.

Among employees who took a follow-up skill check, scores rose by a median of 42%, with reassessment happening an average of 14 days later. Across these workforces, assessed employees met target proficiency in 74.4% of capabilities, and nearly three-quarters of assessed employees have already reached their target.

Two weeks. That is the interval between a measured gap and a measured gain, when development is pointed at something specific rather than distributed evenly across a population.

The reassessment itself is what makes the number defensible. A baseline establishes where a workforce stands on the day it is measured. Only a second measurement answers what an executive committee actually wants to know, which is whether the spend changed anything. Organizations that skip it are left telling a board that people completed the training, which has never been the same claim.

What this looks like in practice

• Baseline what people can do, measured directly rather than inferred from completions, credentials, or self-ratings.
• Direct investment toward verified gaps, which in this dataset means weighting Responsible AI more heavily than the foundational tracks most programs lead with.
• Reassess, so the gain becomes evidence a chief medical officer or a regulator can be shown rather than a claim they are asked to accept.

None of that requires a new function. It requires treating workforce capability as something to be measured rather than assumed.

The question worth sitting with

If a regulator asked tomorrow which of your teams can be trusted to apply AI to a decision that touches a patient, where would the answer come from?

If it traces back to a completion report, you have a record of attendance standing in for proof of judgment. There is a better starting point available.

Life sciences leaders ready to baseline and build Data and AI readiness across their workforce can request a demo.

‍