← Cases
DenmarkHealthcare and life sciences2023 – ongoing

Novo Nordisk

NovoScribe generative AI for clinical study reports

Novo Nordisk's NovoScribe drafts a clinical study report in about 10 minutes instead of 10 to 12 weeks

WorkingVendor's case studyVendor ↗

Problem

Clinical study reports submitted to regulators can run to 300 pages and were written by teams of statisticians, scientists and medical writers; a writer averaged 2.3 reports a year. Each report took roughly 10 to 12 weeks, and the company links every day saved before a drug reaches market to revenue.

Approach

An 11-person internal team built NovoScribe, which combines a trial's statistical outputs with expert-approved text snippets and a document template through retrieval-augmented generation. It used models including Anthropic's Claude on Amazon Bedrock and stored each snippet with its vector embedding and metadata in MongoDB Atlas so that every sentence can be traced to its source. Experts still review the drafts, and the platform was later extended to device verification protocols and patient materials.

Results

-90%writing time for clinical study reportsVendor's case study[1][3]
~10 mintime for the system to generate a report draftVendor's case study[2][1]
30%share of the company's clinical study reports compiled with NovoScribe when MongoDB reported the resultVendor's case study[2]
-95%resources needed for device verification protocolsVendor's case study[1][3]

Takeaway

The 10-minute figure covers draft generation; expert quality review still follows. Suppliers quoting Novo Nordisk staff put the previous drafting time at more than 10 weeks or about 12 weeks.

Categories

Innovation type
Process innovation
AI technology
Generative AI
Evidence
Vendor's case study

Sources

Source summaries

[1]Anthropic2025

Anthropic describes NovoScribe, built with Claude Code, Amazon Bedrock and MongoDB Atlas by an 11-person team. It reports clinical study documentation falling from more than 10 weeks to 10 minutes, a 90% cut in report writing time, a 95% cut in resources for device verification protocols and review cycles roughly halved. It quotes Novo Nordisk's digitalisation strategy director.

[2]MongoDB2024

MongoDB reports that report creation took about 12 weeks with a multidisciplinary team and that the models now generate a draft in about 10 minutes, with quality assurance taking the remaining time. It explains the retrieval design, the use of Claude 3 and other models, and the traceability of each snippet. It states NovoScribe compiled about 30% of the company's reports, with more than 90% expected by year-end.

[3]Unite.AI16 Sep 2026

Unite.AI reports on a September 2026 drug discovery collaboration between Novo Nordisk and Anthropic and, in passing, the existing NovoScribe deployment. It cites Anthropic's figures: documentation time from more than ten weeks to ten minutes, a 90% cut in writing time and 95% fewer resources for device protocols.

Full details are in the original sources.

Added 25 Oct 2026 · Updated 25 Oct 2026
Open case page ↗