From 15 weeks to 10 minutes: Novo Nordisk now drafts clinical study reports with AI. Posts like this are common these days!
What we don’t hear enough about is the rigorous planning and testing that goes into creating a controlled, repeatable solution. There was a lot of groundwork by the Novo Nordisk IT and AI team to power this achievement. Some of it is described in this case study posted by Mongo Vector DB.
Many people who read these headlines imagine a writer typing a prompt into their LLM of choice and generating a document. The reality is very different. The classic iceberg example applies here.
Organizations who want to operationalize their AI solutions will need to:
- Rethink the value stream
- Create quality source content
- Build repeatable solutions
- Enforce security and traceability
What Novo Nordisk Did
Novo Nordisk, a Danish drugmaker, used AI to draft clinical study reports (CSR). Every health authority in the world requires CSRs for regulatory review as part of their determinations of whether a drug is safe and effective enough to approve.
A typical CSR is hundreds of pages long and describes the results of a drug trial. Each CSR integrates data from many sources and content from many authors into a single document. The document undergoes multidisciplinary reviews to ensure the information is accurate and complete.
The CSR is a perfect candidate for AI because it is a summary document with a consistent outline that aligns with guidelines issued by regulatory agencies. Much of the summary information about the trial – the study design, the study rationale, eligibility criteria, and so on – originates in other documents that follow a consistent outline.
Without AI, a medical writer will search, find, copy, paste, and summarize information from source documents into the CSR. This manual effort is time-consuming, taking 10–12 weeks, and prone to human error.
According to this report, with the aid of AI, the Novo Nordisk team was able to reduce the time to 10 minutes of AI effort and 3 days of human effort to validate the results. You can read more about the pharma content revolution here.
Rethink the value stream
A deep understanding of the business drivers is important. Some skeptics will say that the time savings are not as impressive as the new process adds new steps and moves some authoring tasks upstream. There was also a major investment in developing the ecosystem of technology, content, and process that enables AI generation of CSRs. Finally, it took time to set up effective prompts and prime the system to create the CSR.
However, the ability to shift work upstream to deliver a CSR faster is still a huge win. The value driver is earlier submission of the CSR, not overall effort. The ability to efficiently curate all the work done earlier in the process to create the CSR faster meant the CSR was submitted 9 weeks earlier than it would have without the use of AI. Shifting work upstream to bring a drug to market faster saves lives.
The AI value proposition may require us to look at the entire value stream and rethink where work happens. AI may move exploration, analysis, planning, and validation work to the start of a process to support rapid execution later in the process.
In the creation of the AI-generated CSR, we shift human effort from the end of the process to earlier in the process. Work that used to be at the end curating and summarizing information that already exists is shifted to improving the knowledge capture and processing of information.
The expertise of medical writers is shifted to interpret, strategize, narrate, validate, and build AI solutions.
Humans need to focus effort on knowledge capture. The CSR case was possible because the source content and the target content were complete, accurate, and predictable.
Create quality source content
Content is the fuel for AI. How the source content is collected, stored, and managed will have a profound impact on the quality of AI solutions.
Humans are essential to capturing new information—to add AI as a new persona to our content development efforts. The good news is most of the rules that make content findable and understandable by humans also apply to AI.
Basic best practices for writing content for AI include:
- Use consistent terminology
- Provide context through clear structure and metadata
- Write clear, simple sentences
- Ensure clear antecedents
- Beware jargon, colloquialisms, and bias
You can read more on structuring and optimizing content for AI here.
Build repeatable solutions
Once the experimentation phase is over, it is time to operationalize the AI solution. The usefulness of the AI tools depend on the amount of setup and support required to keep the process running smoothly.
The CRS is a great fit because the inputs and outputs of the document are predictable. Novo Nordisk team could build a robust framework of content rules, powerful complex queries, and precision search results that could be applied across similar CSRs.
Enforce security and traceability
Novo Nordisk’s medical writers do not use public tools for CSRs or any other regulatory content. They use only the custom-built tools and vendor-provided applications that are tested and verified by their IT and AI teams.
Building AI solutions will require more than building AI capabilities. Teams will need to build or buy systems to manage the flow of tasks between AI and humans. These workflow tools will need to track content that is AI-generated, human-generated, AI-generated and human-reviewed, or human-generated and AI-reviewed.
Operationalizing AI
Businesses are moving past the experimental phase of AI adoption to create robust, repeatable solutions that transform how work gets done. The Novo Nordisk case demonstrates how a thoughtful, systematic approach to AI implementation can yield remarkable results—not just in time savings, but in fundamental value creation.
The future of human-led AI requires us to reimagine our workflows. It involves shifting human expertise upstream, where critical thinking, creativity, and domain knowledge can have the greatest impact. Human expertise is being redirected away from copy/paste/tweak and handcrafting derivative documents toward higher-value activities like strategy, validation, and information capture.
For organizations looking to follow suit, the path is clear but demanding. Success requires rethinking entire value streams, investing in high-quality structured content, building repeatable frameworks rather than one-off solutions, and establishing robust security and traceability mechanisms. As we move forward, the most successful organizations won’t be those that simply deploy AI tools, but those that thoughtfully integrate human expertise with AI capabilities to create solutions that neither could achieve alone. The result will be faster innovation, better outcomes, and—in cases like Novo Nordisk—getting life-saving treatments to patients who need them more quickly than ever before.
Discover essential capabilities, innovative solutions, and security measures for seamless AI integration. Contact us for a free consultation.
