Want to help your data and analytics teams embrace Agile but don’t know where to start? Wondering why your data team seems to struggle with creating manageable yet valuable stories? Curious why we think Agile for data teams is a distinct challenge?

Data work is often structured more like a pyramid than the familiar “layer cake” metaphor due to the state of data infrastructure technology, common industry practices, and the heavy lift to integrate data before it can be analyzed and visualized. Prevailing Agile wisdom of cutting work into “vertical slices” thus presents significant challenges for Agilists working on data teams! Typical full-stack vertical stories in this environment can easily become too complex, interdependent, and unwieldy to fit into fixed-length sprints. Technical stories can encapsulate smaller work increments but risk becoming too abstracted from the customer’s core problems and trap the team in infrastructure work for too long. An additional impediment to traditional user stories is the highly exploratory nature of advanced analytics and data science projects where in many cases end users lack awareness of what kind of problems can even be solved and technical experts can’t initially predict which solutions will actually be possible.

This session presents successes and lessons learned from applying alternative story decomposition and writing techniques on several data products across multiple teams. Returning to one of the fundamentals of what makes Agile valuable, namely to obtain feedback on feasibility and end user value as quickly and systematically as possible, our approaches strive to ensure teams have small, independent stories while still maintaining a value focus. We discuss ways to decouple the technical stack through stubbing and gradual tightening of the Definition of Done. This technique accommodates the necessary foundational work in the background while also obtaining early feedback about the value of the eventual product delivery options. A second approach incorporates Lean Startup concepts and centers on replacing traditional user stories with testable hypothesis statements that allow for explicit experimentation and risk trade-offs towards relevant milestones such as model quality, performance, predictive reliability, etc. in the context of extreme uncertainty.

Join us as we discuss some of the friction Agilists can encounter on data teams, as well as some validated ideas for meaningful solutions.