Life sciences

Clinical trial data is valuable but slow to use. Novo Nordisk had years of trial data, yet turning it into insight depended on specialists and long workflows; assessing one hypothesis could take months and sometimes ended with the finding that the evidence was too thin.

Data must be harmonized to industry standards before analysis or AI can work across studies, which is a multi-year effort in its own right.

Regulated computerized systems must be validated under GxP rules such as FDA 21 CFR Part 11 and EU GMP Annex 11, and there is no GxP certification a cloud provider can hold on a customer's behalf.¹²

Life sciences is not a separate Microsoft Industry Solution on Learn. Microsoft for Healthcare includes a life sciences data model in preview, while most pharma work builds on Azure data and AI services. Microsoft's GxP guidance lists Azure, Microsoft 365 and Dynamics 365 as in scope and points to qualification guidelines that set out the responsibilities Microsoft and customers share.

Novo Nordisk worked with Microsoft forward deployed engineers on Azure to build a reasoning agent that generates and runs code and performs statistical analysis on its proprietary clinical datasets. The data comes from FounData, which harmonizes more than 200,000 patient-years of trial data to the CDISC, SDTM and ADaM standards; Databricks describes FounData as built on its lakehouse. Biostatisticians and experts validate promising outputs before they influence decisions, and the team ran thousands of automated tests and layered evaluations before production. Novo Nordisk's head of FounData AI applications says the company can now evaluate more than 50 strong ideas a quarter, up from 5 to 10.

Not every question needs an agent. Standard statistical environments and curated data marts answer most routine analyses; an agent helps when many exploratory questions compete for few biostatisticians and the data is already harmonized.¹²³⁴⁵