Drug companies are cutting months off admin tasks with artificial intelligence. The hard part, actually discovering new medicines, remains stubbornly out of reach.
When UBS analysts asked 24 European biopharma management teams how artificial intelligence was changing their businesses this month, they expected to hear about breakthrough drug discovery. What they got instead was a story about paperwork.
Regulatory filings that once took two months now take two days. Annual budgeting cycles that consumed eight months have been compressed to eight weeks. Clinical trial protocols are being translated into 17 languages automatically, and sales teams are receiving AI-generated physician profiles to sharpen their pitches. The productivity gains are real, measurable, and already showing up on income statements. They are just not where the hype machine told investors to look.
The operations story
Across the 24 companies surveyed at the UBS London Healthcare Conference, the clearest near-term AI gains are clustered in the unglamorous work of running a large pharmaceutical business: regulatory documentation, financial planning, medical writing, marketing, and manufacturing.
Genmab's example is striking. The Danish antibody specialist worked with Anthropic to address regulatory concerns about data traceability, and can now structure complex datasets into submission-ready documents in a fraction of the time. Novartis describes a small internal team that transformed its annual budgeting process using AI, reducing both headcount and institutional complexity in the process. Sanofi, meanwhile, is running AI across decades of biosensor data from its flu vaccine production lines, predicting batch failures early enough to intervene, and booking a 3 to 6% yield improvement as a direct cost-of-goods benefit.
These are not speculative future gains. They are happening now and, as UBS notes, effectively extend the commercial life of drugs already on the market by compressing the timelines between development and launch.
Where biology pushes back
The harder question, whether AI can actually generate new medicines, produced more cautious answers.
Management teams were consistent: AI is genuinely useful for identifying potential drug targets and accelerating early chemical design. AstraZeneca PLC (LSE:AZN, NASDAQ:AZN) described using it to find chemical structures faster, reducing reliance on large teams of chemists. Zealand Pharma has built machine-learning models on 27 years of proprietary experimental data to predict where molecules are most likely to fail, helping the company cut losses early rather than late.
But translating a promising target into a drug that works in patients remains deeply difficult, and AI has not solved it. Roche provided the conference's most sobering data point: an AI-identified target for idiopathic pulmonary fibrosis recently failed in phase II, performing worse than placebo. The target identification was correct. The clinical prediction was not. Human biology, the analysts concluded, remains the ultimate bottleneck.
Lundbeck made the same point differently. The company has 20 to 25 AI proof-of-concept projects running, but management was direct that AI will not meaningfully shorten the biological duration of clinical trials. The clock still runs at biology's pace.
What this means for investors
UBS frames pharma as a broad AI beneficiary rather than a sector about to be transformed by a handful of winners. The productivity gains are real but incremental, and the long development cycles that govern drug approval mean any improvements compound slowly. A more dramatic shift would require AI to reliably predict clinical outcomes, not just identify targets, and that capability does not yet exist.
The patent cliff and pricing pressure facing the sector are not going away. What AI offers, for now, is a way to run a leaner operation while waiting for biology to catch up.