The Minimum Viable Analysis (MVA)
What you'll learn: How to resist over-engineering early analyses by starting with the simplest possible approach that could answer your question.
The Core Idea
The Minimum Viable Analysis is about fighting the temptation to build complex models or elaborate pipelines when a simple analysis might suffice. Borrowed from the "Minimum Viable Product" concept in software, MVA means starting with the most basic approach that could plausibly answer your stakeholder's question, then adding complexity only if the simple version proves insufficient.
Think of it like this: if someone asks "Do our customers prefer product A or B?", you don't need deep learning. Start with a simple count or percentage breakdown. Only if that doesn't tell the story clearly do you add statistical tests, segment analyses, or fancier methods.
Why MVA Matters
Many data scientists jump straight to sophisticated techniques—ensemble models, neural networks, elaborate feature engineering—when stakeholders often just need directional insights. Simple analyses:
- Deliver faster: You can answer the question in hours instead of weeks
- Are easier to explain: Stakeholders understand counts, averages, and trends
- Reduce wasted effort: If the simple answer settles the question, you saved days of work
- Expose data issues early: Basic analysis quickly reveals data quality problems
The MVA Workflow
- Ask: "What's the simplest calculation that could inform this decision?"
- Execute that basic analysis
- Share results and gauge whether it's sufficient
- If gaps remain, identify the specific shortcoming
- Add one layer of sophistication to address that gap
- Repeat until the analysis adequately supports the decision
This is iteration in action—you're building understanding incrementally, not all at once.
Key Takeaway: Start simple, deliver fast, and only add complexity when the simple version demonstrably fails to answer the question at hand.