AstraZeneca: Agentic AI Quality-Control System for Pharma Production

Stage 1: From Complexity to Actionable MVP (3 Months)

Task 1: Understanding the Problem & Exploring Opportunities - With this project being sponsored by senior executive stakeholders after a previously unsuccessful version, there was a lot of pressure to get things right. This made forming a concrete understanding of the complex, highly regulated problem space essential in the face of competing approaches and priorities across the team.

To establish alignment and smooth collaboration quickly, I conducted an aggressive research effort, sharing and integrating my findings into developing team goals along the way. This motivated everyone to take a step back, to balance the need for progress with the discipline of understanding the problem space and its scope.

Ensuring Alignment & Shared Understanding: User Personas & Experience Mapping

  • Shared understanding was needed at both the macro systems-thinking level (experience map), and at the micro user-centered level (personas).
  • Combined insights surprised us with the broken nature of the previous solution; creating far more manual work for users with few benefits.
  • The need for completely rethinking this solution was quickly realized, aligning the team on the next step of reconceptualizing the scope of this challenge.

Exploring Potential Solutions: Ideation & Refined System Design Concept

  • A well- understood problem space had the key impact of encouraging everyone to consider new options for a potential solution.
  • I lead the combined team through the ideation process; this allowed us to explore possible solutions while ironing out misaligned goals.
  • Drawing upon SMEs, and my own AI experience, I refined the best aspects of various ideas into a detailed product recommendation.

Task 2: Driving Project Clarity & Initial Design - With an exciting solution concept, it was time to break it down into concrete requirements that could be shared with stakeholders for approval. Our goal was to keep the effort nimble through realistic milestones and tight-knit collaboration. I was delighted by how quickly the team developed a mutual trust, which allowed us to design a successful MVP.

Confidence via Clarity: User-Story Mapping

  • Delivering this solution was a constant team concern due to previous “bottom up” approaches to requirements.
  • To establish clear project goals, I guided this mapping effort, and was surprised by how naturally the team utilized this framework.
  • Utilizing Miro as an online workspace, scope could be negotiated and set, with many technical and design issues accounted for to minimize risk.

Establishing Realistic Expectations: Project Milestones & Resources

  • To gain project approval, stakeholders needed to see these agreed upon project goals in a clear and simple format.
  • Feasible targets with clear timelines and costs were presented to inspire stakeholder investment, and mitigate potential scope-creep.

Trust & Efficiency via Collaborative Design: User-Flows, Wireframes, & Mockups

  • These presented an ongoing opportunity to foster trust and investment through weekly design reviews across disciplines.
  • Artifacts were shared, critiqued, and prompted insights from the team before committing to more detailed deliverables for hand-off.
  • This process maintained a collaborative atmosphere while minimizing expensive rework at the Development/Production level.

Stage 2: Full-Solution Design & Integration at Scale (6 Months)

Task 1: Iterating towards a Full-Solution Design - With a successful MVP demonstrating value, it was time to ramp up the effort to fully flesh out our solution.

As our team scaled up, I had to adapt from a more “hands on” strategic design role to one that prioritized oversight and high-level governance. This presented a significant challenge as I balanced the coordination of the design effort, while folding in and providing clear guidance for new team members.

AI Model Expansion: Specific Considerations & Design Reviews

  • As the system-design became more robust, the solution interface needed to expand to clearly demonstrate these additional functionalities.
  • Scaling the design process, I delegated, guided, and reviewed work for the expanded team at both the system architecture and interface levels.
  • By maintaining a smooth, collaborative workflow, the larger team successfully delivered on more complex functionality ahead of schedule.

Keeping Humans in the Loop: Optimizing Collaborative AI via Prioritized Feedback

  • With a limited window for optimizing the completed solution, the biggest challenge was gaining enough feedback to act upon.
  • Early feedback channels were included in the interface prototype, providing focus for limited usability-testing.
  • This allowed us to diagnose the most pressing issues for users, and iron them out before the new system was deployed.

Task 2: North American Implementation -With our full solution nearing completion, stress-tested, and working beyond expectations, it was time to share it with pharmaceutical production facilities around the country (and later the world).

While we had a robust system, it was critical that on-site personnel implemented it in an optimal way; this would ensure that the QA system would work as intended, yielding the best possible results with a minimum of risk.

Ensuring Smooth & Accurate Scaling: Implementation Guidelines & Resources

  • With our solution nearing completion, guidelines for international integration were needed as this solution was entirely new.
  • Training protocols for AI models, online videos, and troubleshooting resources were designed and shared to ensure that implementations would work as designed for maximum benefit.
  • Combined with scheduled consultations, dispersed production line teams were able to understand and implement the new system quickly and easily.

© 2016 David Service

Version 3.2.0