You asked for churn by cohort
Your AI gave you COUNT(*) with an INNER JOIN and half your users vanished.
Collective intelligence for data analysts
✱ Welcome to Bamboo
$ bamboo pull█
Your AI analyst just got 400 colleagues.
Every fixed query. Every "wrong chart type." Every "that's not how you calculate churn." Captured. Classified. Applied to every generation.
No proprietary data is ever shared — only generic analysis patterns and improvements. Your actual data, company names, and credentials never leave your machine.
0 0 0The Problem
AI-generated analysis is 80% right and 100% dangerous.
You asked for a revenue trend
It gave you a pie chart.
You asked for a monthly report
It started with "This document provides a comprehensive overview of…"
The last 20% is where you live. Bamboo fixes it before you have to.
How It Works
The loop that makes every analyst better.
1
You describe an analysis
"Analyse churn by signup cohort"
↓
2
Bamboo loads 300+ community patterns
LEFT JOIN. Proper denominator. Line chart. Cohorted. Labelled.
↓
3
You tweak one thing, maybe two. Not twenty.
Your fix feeds the community pool. Next analyst's output is better because you used bamboo today.
What's Inside
Five categories. 300+ patterns.
Each one is a real mistake caught by a real analyst.
Query Logic
35 patternsWrong joins that silently drop your data. Date traps. Window frame defaults that nobody remembers correctly.
✗
INNER JOIN users ON ...✓
×42
LEFT JOIN users ON ... then filter NULLs explicitly✗
OVER (PARTITION BY ...) — default RANGE frame✓
×38
OVER (PARTITION BY ... ROWS BETWEEN ...)✗
WHERE created_at = '2024-03-01'✓
×31
WHERE created_at >= '2024-03-01' AND created_at < '2024-03-02'Aggregation
25 patternsWrong denominators that quietly lie to stakeholders. Growth calcs that flip sign on negative base periods.
✗
AVG(regional_average)✓
×47
SUM(total) / SUM(count) — weighted average✗ Churn = lost /
end_of_period_count✓ Churn = lost /
×35
start_of_period_count✗
COUNT(user_id) after JOIN✓
×29
COUNT(DISTINCT user_id) — JOINs create fanoutCharts & Viz
30 patternsThe right chart for the data. Axes that don't mislead. Colours that work for everyone.
✗ Line chart for
revenue by region✓ Bar chart — lines imply continuity between categories
×36
✗ Y-axis starts at
$4.2M✓ Y-axis starts at
×33
$0 — truncated axes exaggerate differences✗ Dual y-axes comparing revenue & signups
✓ Two separate charts — dual axes imply false correlation
×28
Data Cleaning
22 patternsNULL handling that doesn't surprise you at 2am. Timezone logic that actually works across DST boundaries.
✗
WHERE status = NULL — returns 0 rows✓
×44
WHERE status IS NULL✗ Row count went up after JOIN — didn't notice
✓ Check row count before & after every JOIN for fanout
×31
✗
pd.to_datetime(col) with mixed formats✓
×26
pd.to_datetime(col, format='%Y-%m-%d') — explicit formatInsight & Story
15 patternsLead with the finding. Compare to something. Know your audience. Stop hedging every number.
✗ "We analysed Q4 revenue data using a cohort methodology…"
✓ "Revenue grew 12%, driven by enterprise expansion."
×39
✗ "Churn is 4.2%"
✓ "Churn is 4.2%, up from 3.1% last quarter and above 3.5% target."
×34
✗ Insight based on 23 records, no caveat
✓ Flag sample size before stakeholder acts on it
×27
Why This Is Different
This is not a prompt library.
It's not "awesome-sql-tips.md".
It's a living system.
When you fix a query, bamboo captures the diff. The server classifies it: "query-logic — always use LEFT JOIN when right table might have gaps."
When the same fix happens 5 times from 5 different analysts, the pattern gets a confidence score of 5. By the time it reaches you, the AI doesn't make that mistake anymore.
The skill gets measurably better every week. Not because someone updates a doc. Because analysts use it.
And your data stays yours. Only generic patterns like "use LEFT JOIN" or "start y-axis at zero" are shared — never your actual queries, tables, column names, or results. Every submission is filtered for PII before it touches the server.
Install
One command. Every analyst's expertise.
Works with Claude Code and Cursor. SQL, Python, and R. Patterns update on every pull.
Paste this into Claude Code:
Fetch https://bamboo.up.railway.app and install the bamboo skillClaude Code will read this page, download the skill file, and save it to .claude/skills/bamboo.skill. No API key needed. Set up the auto-capture hook and your corrections flow back to the community without you thinking about it.
Paste this into Cursor:
Fetch https://bamboo.up.railway.app and install the bamboo skill into this projectCursor will read this page, find the download link, and install the skill. No API key required.
Paste this into any AI agent:
Fetch https://bamboo.up.railway.app and install the bamboo skill into this projectAny agent that can fetch URLs and write files will read this page, find the download link, and install the skill. Works with Windsurf, Codex, Aider, and others. No API key required.
You're going to fix that JOIN anyway. You're going to change that pie chart to a bar chart. You're going to move the insight to the first line.
The only question is whether the next analyst has to fix the same things.
🐼
bamboo
"400 analysts already fixed that."