Novo Nordisk Scales to 2,500+ Use Cases with Secure Generative AI Using Amazon Bedrock

Novo Nordisk produces therapies for chronic diseases such as diabetes, obesity, and rare blood disorders. The company has been on an AI journey for several years, and it began experimenting with large language models and generative AI in 2023. Instead of trying to predict what would stick, Novo Nordisk wanted to provide an environment in which employees could innovate with generative AI for their use cases in nonregulated processes while continuing to do secure, compliant, and cost-effective work. “A lot of colleagues had great ideas for how to solve their own business problems using generative AI, but they were limited by the fact that they couldn’t develop the applications or maintain the infrastructure to support them,” says Jepsen. “We also found it difficult to accurately predict which of these use cases would take off and which ones were less valuable.”

On AWS, Novo Nordisk created a self-service platform for its employees so that they could use the foundation models in Amazon Bedrock to build and customize chatbots for specific use cases. To implement the solution, the company worked with Cloud2 Oy (previously as KeyCore), an AWS Partner, to validate its architectural designs so that the solution would be secure and scalable. “Staying within our AWS environment alleviates a lot of the headaches around security,” says Jepsen. “Having this foundation with generative AI capabilities makes very rapid innovation possible, and it significantly reduces both the risks and the cost around innovation.”

Teams across Novo Nordisk use the platform, which is centered around chatbots and agents, to increase productivity, standardize workflows, and improve information quality. “It’s been great to see all the things that our colleagues are inventing using our platform,” says Jepsen. “People have been quick to adopt it and happy to experiment with it. The democratization of innovation has been the biggest outcome.”

Novo Nordisk teams can use the platform to create a chatbot from scratch in days rather than months. Teams can use the newest foundation models as soon as they are released in Amazon Bedrock, testing and benchmarking models based on what’s best for the use case and building with agility as the technology evolves. “The tech is transformative,” says Jepsen. “It’s been so great to see all the innovation that the technology supports.”

Using the generative AI platform, over 25,000 employees have created chatbots for over 2,500 unique use cases, such as retrieving information, drafting documents, or acting as a virtual colleague or critic. For example, employees receive help drafting documents, where the chatbots can correct common mistakes and formatting issues so that reviewers can focus on substantive feedback and value-added tasks. Users not only can create a chatbot but also share it with their teams so that everyone has access and can iterate further.

Similarly, a range of information-retrieval chatbots are helping employees find the information that they need quickly from various source documents, saving time. More than 1,000 employees use the company’s general-purpose, off-the-shelf chatbot. It processes over 26,000 prompts in a month, moving from a proof of concept to a minimum viable product in 1 month and training its largest use case on 140,000 documents. “Using the chatbot helps employees be more informed as they discuss information,” says Jepsen. “This technology improves the quality of work and removes the overdependence on specific employees in the process change with particular knowledge.”

Novo Nordisk keeps costs at a minimum by running the solution on serverless AWS services, including Amazon DynamoDB—a serverless, NoSQL, fully managed database—and AWS Lambda, a serverless compute service. “The important thing for us is that it is able to scale,” says Jepsen. “We can’t predict which use cases will catch on, so we want our people to build things on AWS with minimal base costs.”

Each use case costs around 10 dollars per month to run on AWS. “We are reducing cost significantly, but we’re more interested in how we are adding value,” says Jepsen. “Having this foundation through Amazon Bedrock and our platform supports very rapid innovation. We are bringing the cycle of innovation for a use case down from several months to a couple of days, or just 1 hour for a proof of concept.”

Novo Nordisk views generative AI primarily as a catalyst for innovation and enhanced value creation, not as a means to cut costs. The company aims to liberate its employees from routine tasks through the technology so that they can concentrate on more fulfilling and impactful work, which boosts job satisfaction. “Our implementation of generative AI isn’t about replacing jobs,” says Jepsen. “Rather, it’s about significantly elevating the quality of work and enabling our team to focus on what truly matters.”

Novo Nordisk will continue using generative AI and related technologies to improve its processes and workflows, with an eye toward compliance and risk management. The company is also looking at using Amazon Neptune, a serverless graph database, to build an extensive knowledge graph. “Ultimately, we hope to use this technology so that we can provide even more value to our patients,” says Jepsen.

For other companies looking to innovate, Jepsen has this advice: “Don’t try to predict or dictate what innovation will look like. Instead, help your organization in using generative AI technology safely and securely.”