Quality Copilot: Transforming Workflows at Novo Nordisk Through AI and User-Centered Design

Senior UX/UI Designer, 2025 Novo Nordisk & Microsoft

Duration: 3 month

Please note that to ensure the non-disclosure of any internal or confidential information, this case study provides general insights into the product and process rather than specific details. Additionally, the layout includes low-fidelity data placeholders to prevent the dissemination of sensitive information.

Context

To ensure the safe production and delivery of medicine to Novo Nordisk patients, continuous environmental monitoring in production facilities is essential. This involves a dedicated team of experts who assess the environment, take samples, and analyze data for contamination control. In cases of excursions, Environmental Monitoring (EM) Supporters must quickly determine the safety of the medicine to avoid prolonged quarantine.

The Quality Copilot project supports EM Supporters in documenting deviations and utilizes emerging AI capabilities in environmental monitoring.

Designing AI Solutions with Compliance in Mind

Designing a platform that utilizes AI solutions follows a similar process to creating any other product, emphasizing a user-centric approach. It begins with understanding the problem space through research, designing solutions, and continuously testing and iterating.

However, developing an AI solution in a regulated environment—particularly one that monitors the safety of medicine offered to patients—presents a unique set of challenges.

AI technologies provide significant opportunities to enhance efficiency and productivity while addressing gaps in workflows. But how can we ensure these technologies are effectively integrated while maintaining compliance and safety standards? These were the challenges I had to address during the design phase.

““How might we design a solution where AI acts as a collaborative partner that enhances productivity while ensuring humans remain actively in the loop?”
“How might we leverage AI to optimize workflows while ensuring the safety and integrity of the environment?”

Many products include AI chat interfaces, expecting users to interact and ask questions from an AI chatbot. However, this approach requires users to adjust their focus, understand what the models are capable of and how to leverage it, create clear prompts, and determine how to use the responses in their tasks.

Designers need to acknowledge that users are not always proficient in crafting prompts.

The Quality Copilot sought to challenge the conventional practice of merely embedding AI chat features into existing products, which aims to address all user problems. Instead, we took the time to thoroughly understand the entire user journey of the EM deviation handling workflow and designed AI features in true collaboration with the users they serve. Our goal was to develop a comprehensive new platform optimized for this workflow, while ensuring scalability to accommodate the diverse needs and objectives of users in the future.

Designing AI Solutions: Expanding Beyond Chatbots

This screenshot shows the onboarding flow and key features of the MVP for the Quality Copilot, designed using Figma.

If you need further adjustments or alternatives, feel free to ask!

MVP Scope of the Quality Copilot

Changing Data Parameters and Downloading Data

The EMS can modify the default data parameters associated with the deviation to evaluate a broader range of data and download it for inclusion in the required reports.

Present Relevant Data

The relevant data required to close the deviation is automatically gathered and presented to the EMS upon opening the deviation case, eliminating the need to access multiple systems to retrieve the information.

Descriptive AI Summary, without the need to prompt the AI

To release the EMS from describing the data into text, AI Generated analysis available from the data to help the deviation handling process.

MVP Quality Copilot Prototype

This screen recording showcases the main features of the MVP version of the Quality Copilot. Although the data is presented as low-fidelity outlines and is not visible, the interactions are clearly observable.

The UX of AI.

  • Validation and References

    Enabling users to view and manage the sources of AI responses is becoming a standard practice. In the design of the Quality Copilot, it was essential that the data is presented alongside the AI-generated data summary, allowing users to easily cross-check and validate the content.

  • Rating

    Empower users to report AI risk concerns, including privacy, security, toxicity, bias, or incorrect information in the AI-generated responses.

    A thumbs-up or thumbs-down allows users to communicate to engineers the effectiveness of the model's design.

  • Tone of Voice

    For the Quality Copilot, defining a clear tone of voice was crucial to instill confidence in users.

    Users expressed, it is vital that every line of the AI summary is accurate, based on data, free from assumptions and unnecessary details, and expressed concisely. To fine-tune the tone of voice of our AI output, the content underwent multiple rounds of user feedback and close collaboration between UX and data engineers to ensure consistent and professional quality.

  • Caveat

    It's essential to inform users about the shortcomings or risks associated with the model or the technology as a whole. Caveats can be invaluable, particularly for less technical users, as they serve as straightforward signals of the product’s limitations. Here’s a different example using an everyday item:

    In this sense, it is similar to a warning label on a microwave oven that advises users not to use metal containers.

  • Visual cues to help users identify AI features or content

    How do you know if you are interacting with validated data or data generated by AI? Color has emerged as a helpful signal to help users identify AI products and features. There is by no means a set color scheme, however rainbow and gradient became an emerging trend.

  • Symbols

    Symbols are images that represent AI within the interface. They serve the same purpose as color differentiation but also accommodate individuals who are color blind, enhancing visual understanding for all users.

Final Summary

In the rapidly evolving landscape of artificial intelligence and user-centric design, the Quality Copilot project undertaken at Novo Nordisk stands as a paradigm of innovation and compliance. Over the span of three months, our efforts focused on creating a sophisticated platform that leverages AI to enhance workflows in a regulated environment, ensuring that patient safety and data integrity remained at the forefront of our design process.

Furthermore, we established a refined tone of voice for the AI output, ensuring clarity and precision while maintaining user trust. The strategic use of visual cues and symbols facilitated user interaction, making it easier to distinguish between validated data and AI-generated content.

In conclusion, the Quality Copilot not only serves as a powerful tool for enhancing productivity but also represents a commitment to robust compliance and safety standards within the pharmaceutical industry. This project sets a new benchmark in the utilization of AI, illustrating how thoughtful design and collaborative user engagement can lead to innovative solutions that address complex challenges in healthcare.