Can Agile methodologies help improve data quality in AI?
We are trying to apply Scrum to our machine learning workflow, but the unpredictable nature of data quality issues makes it hard to plan sprints. A single data corruption issue can blow up a two-week sprint. Has anyone successfully adapted Agile to handle the "unknown unknowns" of data preparation in AI projects? Any specific ceremonies that helped?
2025-11-21 in Agile and Scrum by Kevin Foster
| 11068 Views
All answers to this question.
Agile can work, but you have to treat data discovery as its own "Enabler" or "Spike" in your backlog. In my 2023 projects, we never committed to a modeling story until the data quality spike was completed. This allowed us to identify bottlenecks before the sprint even started. We also added a "Definition of Ready" that specifically required a data health report. It prevents the team from over-committing to outcomes that the current data simply cannot support. It’s all about visibility and adjusting your velocity expectations based on the state of the data.
Answered 2025-11-23 by Heather Collins
Heather, how did you handle the stakeholders who felt that these "data spikes" were slowing down the delivery of actual features?
Answered 2025-11-25 by Joseph Reed
-
Joseph, I had to frame it as a risk mitigation strategy. I showed them that a failed sprint due to bad data is much more expensive than a dedicated week of discovery.
Commented 2025-11-26 by Charles Baker
We found that having a dedicated "Data Custodian" attend the Daily Standup helped us catch data quality issues much faster than waiting for the review.
Answered 2025-11-27 by Sandra Martinez
-
That’s a great idea, Sandra. Having that direct line of communication between the data source and the dev team is vital in an Agile environment.
Commented 2025-11-28 by Kevin Foster
Write a Comment
Your email address will not be published. Required fields are marked (*)

