|
Welcome back! In the ultracompetitive world of drug development, getting a new medication on the market as fast as possible can have major business implications. There’s perhaps no better recent example than Novo Nordisk, which has generated nearly $100 billion in sales of weight loss drugs Ozempic and Wegovy while trying to fend off competition from Eli Lilly. Now, Novo Nordisk says it’s using AI agents to shorten by weeks or months the time it takes to both start new clinical trials and to complete them. The new tools are expected to add tens or hundreds of millions of dollars in additional revenue while helping the company hire fewer employees, according to Chief Digital Officer Stephanie Bova. The company already was using AI models like Anthropic’s Claude to speed up drafting documents submitted to regulators for drug approval, as we previously wrote. But recently, Novo created more advanced AI to audit and coordinate the clinical drug trials themselves, she said. To train the agents, the drugmaker used software from the AI startup Celonis to pull Novo’s internal data, as well as information disclosed by competitors about how to run clinical studies. (Celonis, which is headquartered in Germany, has raised $1.6 billion from backers including Silver Lake and Accel.) One agent Novo developed using the Celonis software, which relies on AI models from Anthropic, OpenAI, and other providers, helps employees manage clinical trials. For instance, the agent detects possible risks that could cause trials to take longer, such as a step in protocol that researchers missed or data that looks incomplete. It can automatically send a Microsoft Teams message to leaders overseeing the trial to suggest steps to avoid the delay. (Celonis helps Novo figure out which provider’s model to use for the agent’s tasks.) Novo has been developing such AI agents for about a year, and Bova said they already appear capable of automating parts of the clinical trial process that would have previously been done by humans, such as choosing the best country or location for a clinical trial based on patient availability and local regulations. “We‘re looking at shortening the time from first in human trials to when we submit the drug for approval,” Bova said. “Thats the big question that all drug development leaders have in mind, is how long is this going to take—it’s a race.” Bova declined to say how much Novo Nordisk spends to run the new AI, but said the savings greatly outweigh the cost. Bova estimates that, for new trials starting in the coming year, the AI agents will help shave months or possibly years off the time it takes to get new drugs approved. Pharmaceutical firms have long been at the forefront of using advanced machine learning for core research, so the industry is particularly well positioned to harness the latest AI. Bova, who previously worked at drugmaker Novartis from 2019 until 2022, says she believes competitors are developing similar AI agents. “A week of time saved in getting to market can mean tens of millions if not hundreds of millions of dollars in peak revenue impact,” Bova said. “So as long as we’re shortening that timeline, I wouldn't even bother doing a business case calculation because the juice is clearly worth the squeeze.” Finding Redundancies Bova said she expects the drugmaker to save money on headcount by needing fewer human employees in the future—for instance, an AI agent analyzing the company’s business processes noticed that three different departments were all overseeing the drafting of regulatory documents. The agent has also started to handle data analysis that Novo Nordisk historically outsourced to several hundred contractors, she said. Still, that doesn’t mean all its AI efforts have been worth the money. For instance, Bova said the company previously built an AI tool, Found Data, that its researchers used to pore over decades of prior clinical trial data to look for potential trends past studies had missed. But Novo “scaled back” its use of the tool, which is powered by Anthropic’s Claude model, because it was expensive to run and didn’t seem to lead to many noticeable advances, she said. “[T]here’s been this mad dash to have AI use cases without thinking about whether it's the best use of your time and money to do it that way,” she said. “If I can do it better and cheaper and more reliably in Excel, I’m going to tell you to stay in Excel.”
| 
