Prioritization as code: An AI-supported framework for platform engineering (Eleanor Millman and Mina Tawadrous)

Eleanor Millman and Mina Tawadrous introduce the episode, sharing their roles at SiriusXM and the context for their work. They discuss how SiriusXM built a large-scale internal developer platform supporting 700,000 developers across various languages and tools, with diverse needs and skill levels. The challenges they faced in prioritizing features and projects without traditional metrics or a direct revenue link are highlighted, setting the stage for their innovative approach.

The speakers detail how they devised a prioritization formula inspired by RICE, incorporating impact, urgency, effort, and additional factors like internal stakeholder needs. They explain how impact is broken down into seven specific categories—development speed, runtime reliability, security, cost reduction, platform engineering efficiency, user trust, and data enablement. The impact factors are weighted and can be adjusted based on company priorities, with dynamic reassessment over time driven by developer feedback.

The podcast explores how SiriusXM scores projects using impact factor ratings (0, 1, 2, 3, 5), incorporating user data and surveys to make ratings more objective. They emphasize the importance of documenting assumptions behind scores, treating them like code, and maintaining transparency through version control. Iterations on the model are driven by changing organizational priorities and data insights, making project prioritization more aligned and less emotionally charged.

They discuss difficulties in reconciling conflicting assumptions between teams, illustrating how biases like the 'curse of knowledge' can impair accurate scoring. The concept of storing assumptions as code—versioned, visible, and timestamped—is introduced as a way to resolve assumption conflicts and improve decision-making. This approach shifts debates from emotional disagreements to fact-based discussions, fostering a shared understanding.

The episode introduces AI integration as a tool to surface relevant assumptions and past knowledge, reducing manual effort and improving recall. AI aids in connecting initiatives with existing data, assumptions, and company goals, while humans remain in the decision-making loop for validation. The use of AI enhances the scoring process, making it faster and more reliable, and helps maintain an up-to-date, collaborative assumptions repository.

Eleanor and Mina outline three steps for organizations to adopt a similar framework: define impact impact factors and weights aligned with company goals; score projects by documenting and reviewing assumptions; and integrate AI to improve the recall and validation of assumptions. They encourage continuous iteration and stakeholder collaboration to refine the model over time.

The hosts answer questions from the audience, covering how to evaluate projects with potentially enormous costs, manage conflicts when the framework prioritizes differently from stakeholder desires, and extend the approach beyond developers to support non-technical users. They highlight the importance of transparent, data-driven conversations, and the flexibility to adapt impact factors for broader organizational needs.