Course contentsShow
← Data Science
Lesson 2115 of 2,14548. Data Science in PracticeFree lesson

Knowing When an Iteration is Complete

Criteria for deciding when additional iterations provide diminishing returns and it's time to move forward.

Knowing When an Iteration is Complete

What you'll learn: How to recognize when additional iteration cycles provide diminishing returns and it's time to move forward in your data science project.

The Challenge of Iteration Boundaries

Data science work is inherently iterative (lesson 2109), but every iteration has a cost: time, resources, and opportunity cost of not moving forward. The hardest part isn't doing another iteration—it's knowing when to stop iterating and ship your work.

Criteria for Completion

An iteration is complete when one or more of these conditions are met:

1. Success metrics are satisfied — You've hit the performance thresholds defined during problem scoping (lesson 2094). If you agreed a model needed 85% accuracy and you're at 86%, additional tuning may not be worth it.

2. Stakeholder needs are met — The fast feedback loops you established (lesson 2111) confirm your current solution answers their questions adequately. Perfect is the enemy of done.

3. Marginal gains become small — You spent a week on the previous iteration gaining 3% accuracy. This iteration, after similar effort, you gained 0.5%. The curve is flattening.

4. The timebox expires — Your sprint planning (lesson 2113) allocated two weeks for this phase. Time's up, and you have something defensible to show.

5. Remaining improvements require unavailable resources — Further gains need data you can't access, expertise you don't have, or computational resources beyond budget.

The MVA Test

Compare your current work against your Minimum Viable Analysis (lesson 2110). If you've substantially exceeded it and stakeholders are satisfied, you're done. Don't gold-plate.

Think of it like editing a document: At some point, you're no longer making it better—you're just making it different. That's your signal to ship.

Key Takeaway: An iteration is complete when you've met success criteria, satisfied stakeholder needs, hit diminishing returns, exhausted your timebox, or lack resources for meaningful improvement—whichever comes first.