Application of Bayesian Belief Network for Agile Kanban Backlog Estimation
This paper presents an approach based on influence diagrams for reducing uncertainty in Agile Kanban backlog feature completion time. Agile project management techniques, including SCRUM and Kanban, are prevalent in software development and spreading to other product development fields. A key artifact of Agile is the product backlog, containing work which needs to be completed by the development team. Internal and external stakeholders often require projections for completion of backlogged requests or features. Current estimation techniques such as duration assignments through planning poker and the use of story points to calculate velocity require persistent team input, while task counting has limited accuracy. Therefore, an influence diagram (also known as a Bayesian belief network) was generated to probabilistically assess factors influencing the completion time of backlog items. Statistical functions and uncertainty nodes were validated through data collected from a product development team practicing Agile Kanban. In addition to lowering the barrier to adopting backlog estimation, this model accounts for factors influencing lead time that current techniques disregard such as re-prioritization and feature or request additions. This approach can provide a simpler, more robust representation of project backlog while effectively using team resources.
This proceeding was published as Weflen, Eric, Kevin Korniejczuk, Sharon Lau, Steve Kryk, Cameron MacKenzie, and Iris V. Rivero. "Application of Bayesian Belief Network for Agile Kanban Backlog Estimation." In Proceedings of the 2018 IISE Annual Conference. K. Barker, D. Berry, C. Rainwater, eds. (2018). Posted with permission.