How to Design Analysis in Claude Code

Most analysts jump straight into SQL when a stakeholder says "conversion dropped." That's how you waste two weeks on the wrong analysis. This session walks through a 7-line framework that prevents that, and a Claude Code system that automates the whole process.

Sravya Madipalli Tangeda (Senior Manager of Data Science, Superhuman) walks through a 7-step analysis design framework that forces the right decisions before you write a single query. Then she demos the full system in Claude Code: one sentence in, structured analysis plan out.

Shane Butler and Hai join for Q&A on connecting Claude Code to your company context, eliminating BI tools, and how the data role is changing.

What you'll see:
- The 7-step analysis design brief (question, decision, hypothesis, comparison, segments, confounds, criteria)
- Live Claude Code demo: from a PM's hunch to a shareable V1 analysis plan
- How the system generates hypothesis trees, confound scans, and investigation priorities
- V2 iteration with stakeholder feedback incorporated
- Google Doc export in real time

0:00 Welcome + intros
4:05 Quiz: Your VP says conversion dropped. What do you do?
6:22 Why clarifying questions come first
7:35 What you'll walk away with
8:22 Quick demo: one sentence to Claude Code
10:51 What the system produced
14:16 The 7-line analysis design framework
18:10 Bad vs. good: Question and Decision
21:09 Bad vs. good: Hypothesis, Comparison, Segments
23:11 Bad vs. good: Confounds and Criteria
24:23 Every failed analysis breaks at least one line
26:31 System architecture preview
29:01 Full demo: analysis design pipeline in Claude Code
38:18 Q&A: How much does Claude need to know about your company?
43:21 The V1 analysis plan (Google Doc output)
44:37 V2: Incorporating stakeholder feedback
46:23 Architecture recap
47:51 Q&A: Data visualization and dashboarding
51:55 Q&A: Eliminating BI tools
54:52 Q&A: Feedback loops and insight-to-action
56:21 The future of the data role
61:13 Q&A: Career advice for entry-level analysts
66:51 Closing

LINKS
AI Analyst repo (open source): https://github.com/ai-analyst-lab/ai-analyst
Slack community: https://bit.ly/ai-connect
AI Analytics for Builders course: https://maven.com/dataneighbor/ai-analytics-for-builders
Claude Code bootcamp: https://maven.com/dataneighbor/build-ai-analysts-in-claude-code

Presented by AI Analyst Lab (Shane Butler, Sravya Madipalli Tangeda, Hai)

Видео How to Design Analysis in Claude Code канала Data Neighbor Podcast