From 15 weeks to 10 minutes: Novo Nordisk now drafts clinical study reports—hundreds of pages each—with AI. What once took 50+ writers now requires just three, plus AI (in this case, Anthropic's Claude). That's a 94% headcount reduction and 92% cost-savings (because all that Claude use costs one writer's salary). The AI doesn’t work alone. Humans oversee, refine, and guide it. One cool thing they're doing: using retrieval-augmented generation to save and retrieve human-approved content, ensuring once-reviewed definitions and insights don’t need rechecking. I’d be skeptical, but I’ve seen similar AI-driven efficiency gains, with no loss in quality (and sometimes improvements), in blinded head-to-heads we've run of content generated by AI, humans, and AI with human oversight. This is the future: AI + human oversight = faster, cheaper, and just as accurate. Also the future: Novo isn’t firing writers—it’s just hiring fewer, and shifting investment to other departments.
Hailong Wang
6mo
It is happening and improving for Gen AI in enterprise use (AI RAG, AI Agents, and Agentic AI platforms). The key is domain knowledge implementation to refine and shape planning and reasoning guardrails, as well as collaboration and orchestration with AI, guided by HITL oversight and approval.
Doug Barlow
6mo
The problem I see today is that many people show blind trust in AI. AI still hallucinates and states fake/false information as fact. Sure, I can now generate legal docs in seconds that would have cost $100s or $1000s to bill out. But unless you review everything AI generated with a very fined-tooth comb, you might miss a "surprise." Maybe think of AI as a very-fast intern. Quick output with questionable outcomes. All work would need to be verified by humans and then the intern gets some feedback and generates the work again. Very iterative. Sure, it will be awesome as AI improves and it will cause major shifts in many industries. But we are years away from that, despite what the marketers are saying.
This is exactly what AI augmentation should look like—enhancing efficiency, reducing bottlenecks, and allowing experts to focus on higher-value tasks, rather than repetitive manual work. The key takeaway here? AI isn’t replacing human expertise—it’s amplifying it. Retrieval-augmented generation (RAG) ensures consistency, while human oversight maintains accuracy. The companies that master this balance will outpace their competitors in both speed and innovation. Curious—how long before this level of AI-driven efficiency becomes the industry standard?
Gabor Fari
6mo
I happen to know a bit about the topic. Yes, AI can literally generate the drafts in minutes. And yes, RAG plays an important role (GraphRAG is even better). However, what the hyped article conveniently omits is that it takes months to set up the CSR, and then it is first run with simulated data, to test the system, and the prompts. Then, at the end of the trial, after the database lock, the GenAI is run again, and the output is checked. So, what basically happens is that the timeline is moved from after the database lock to before. This is a tremendous saving and a huge productivity improvement thanks to GenAI. There is meticulous planning involved, and the medical writers spend the time up front to plan, and to check the output. It's misleading to state that it only takes 3 medical writers and not 50, because the time allocation is different, and medical writers spend more time up front.
Gaurab Jayn
6mo
Very informative, Simon. Kudos to Novo Nordisk on this achievement! AI is making a noticeable impact across biopharma, not just as a buzzword, but as a practical tool that's helping researchers bring new treatments to market faster, improve the way clinical trials are run and predict how patients might respond to therapies. By handling large volumes of complex data,AI is becoming an essential part of driving innovation. As I can experience, in everyday healthcare, AI is now being used to gather and interpret patient data, keep track of medical histories, and create more personalized treatment plans and dietary recommendations. This is helping people manage chronic conditions better like diabetes and weight mgmt., and focus on prevention. Human intervention is still at the heart of all of this. Experts remain essential to validate what AI delivers. But there’s no doubt AI is now doing a lot of the heavy lifting behind the scenes.
Not surprised at all. Last weekend I supervised ChatGPT/Perplexity in conducting deep research with very specific instructions to create a narrative review that I could assess for quality. In just one evening, it produced something that could probably be published.
It in the mundane re-creation or compiling of data that our LLMs can really be efficient in generating the initial drafts. Might take a person some days, vs a few minutes. Then the editing and review can begin much sooner, which is still the same QC as if the drafty was from a person.
Global Head GenAI & AI Agents @Sanofi | Digital Transformation Leader | Healthcare • Beauty • Consumer Tech
6moIt’s interesting to see Novo Nordisk embrace this industry trend—it highlights just how transformative AI can be in structured, process-driven areas like clinical studies reporting. This use case, where strict guidelines and validation frameworks already exist, are naturally the "low-hanging fruit" for scalable AI adoption. The real challenge—and opportunity—lies in applying AI to less structured, more creative, and human-driven functions within organizations. That’s where the path forward becomes less straightforward, but potentially even more impactful. As for “faster and cheaper,” I’d argue that’s a near-term consequence of AI + human oversight, not its ultimate future. Framing AI’s potential solely around efficiency is like saying electricity is just about lighting a room. The real story will unfold in how this technology reshapes industries, business models, and human potential in ways we can’t yet fully predict.