How to Perform Data Analytics with PM Skills: A Complete Guide for Product Managers

PM Skills provides a zero-code data analytics plugin that lets product managers generate SQL queries, run cohort analyses, and evaluate A/B tests through simple slash commands in Claude or compatible AI assistants.

The pm-skills repository by phuryn ships a specialized data-analytics plugin designed specifically for product managers who need actionable insights without writing code. This skill-first architecture exposes three core analytical capabilities—SQL generation, cohort analysis, and A/B test evaluation—through intuitive commands that work across Claude, Gemini, OpenCode, and other AI assistants. This guide walks you through how to perform data analytics with PM Skills using the actual source files and command structures found in the repository.

Core Capabilities of the PM Skills Data Analytics Plugin

The data analytics functionality in PM Skills centers on three markdown-based skills stored in the pm-data-analytics/skills/ directory. Each skill encapsulates a complete analytical workflow, from data ingestion to actionable recommendations.

SQL Query Generation for Multiple Dialects

The SQL query skill at pm-data-analytics/skills/sql-queries/SKILL.md transforms natural language requests into production-ready SQL for BigQuery, PostgreSQL, MySQL, Snowflake, and other dialects. When you invoke /write-query, the skill parses your schema file, maps table relationships, and generates a dialect-specific query with inline comments, performance notes, and optional test scripts.

Cohort Analysis and Retention Metrics

The cohort analysis skill at pm-data-analytics/skills/cohort-analysis/SKILL.md computes retention curves, feature-adoption trends, and segment-level insights from uploaded CSV, Excel, or JSON files. The /analyze-cohorts command validates your data structure, calculates retention rates, generates heat-map visualizations, and outputs reproducible Python pandas snippets for further exploration.

A/B Test Statistical Evaluation

The A/B test analysis skill at pm-data-analytics/skills/ab-test-analysis/SKILL.md evaluates statistical significance, validates sample size adequacy, computes confidence intervals, and issues ship/extend/stop recommendations. The /analyze-test command accepts CSV test results and runs proportion tests (Chi-square) or mean-difference tests (t-test, Bayesian alternatives), delivering plain-English verdicts with concrete next steps.

How to Perform Data Analytics with PM Skills: Three Workflows

All three capabilities follow a consistent skill-first architecture. The command wrappers in pm-data-analytics/commands/ handle prompting and file orchestration, while the markdown skill files contain the core analytical logic. This separation makes the functionality portable across AI assistants.

Workflow 1: Generate Production SQL with /write-query

The command wrapper at pm-data-analytics/commands/write-query.md defines the syntax and workflow for SQL generation. You can upload a schema file (DDL, diagram description, or plain-text table list) or describe your tables conversationally.

/write-query Show me monthly active users for the last 30 days, broken down by plan tier (BigQuery)

The skill processes this through a multi-step pipeline: understand the metrics request, parse the schema using simple heuristics (regular expressions for table/column extraction), generate the dialect-specific query, and explain the assumptions. The output includes performance notes—such as partitioning recommendations—and optional test scripts.

-- Monthly Active Users by Plan Tier – last 30 days
WITH events AS (
  SELECT
    user_id,
    plan_tier,
    DATE(event_timestamp) AS event_date
  FROM `my_project.analytics.events`
  WHERE event_timestamp BETWEEN TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 30 DAY)
    AND CURRENT_TIMESTAMP()
)
SELECT
  plan_tier,
  COUNT(DISTINCT user_id) AS mau
FROM events
GROUP BY plan_tier
ORDER BY mau DESC;

The generated query identifies the events table, infers the plan_tier column, and structures a CTE for date filtering. Note that no external SQL parsing libraries are required—the skill uses regex-based heuristics built directly into the markdown logic.

Workflow 2: Run Cohort Retention Analysis with /analyze-cohorts

The cohort command at pm-data-analytics/commands/analyze-cohorts.md triggers the skill for retention analysis. Upload a CSV with columns like cohort_month, user_id, weeks_active, and engagement_score.

/analyze-cohorts

The skill validates data integrity, calculates retention rates by cohort, and generates visualizations (heat-map matrices and line charts). It also outputs a complete Python script for reproducibility.

import pandas as pd
df = pd.read_csv('cohort_engagement.csv')

# Pivot to cohort × week matrix

retention = df.pivot_table(
    index='cohort_month', columns='weeks_active',
    values='user_id', aggfunc='nunique')
retention = retention.divide(retention.iloc[:,0], axis=0)

Sample output: 4 cohorts (Jan-Apr 2025) with 1,200+ users each, showing Week 4 retention declining from 65% to 55%. The skill recommends investigating the April drop-off and suggests follow-up user interviews.

Workflow 3: Evaluate A/B Tests with /analyze-test

The A/B test command at pm-data-analytics/commands/analyze-test.md accepts CSV files with variant, conversions, and impressions columns. It runs statistical tests using standard formulas (Z-score, pooled variance, Bayesian posterior) documented inline in the skill file.

/analyze-test

Output includes the lift percentage, p-value, confidence interval, and Bayesian posterior probability. Example result: Variant B lifts conversion by 3.2% (p = 0.012, 95% CI [0.5%, 5.9%]), with 98% Bayesian probability of improvement. Recommendation: Ship the new checkout flow and monitor week-over-week stability.

Key Integration Points and Extensibility

Understanding how to perform data analytics with PM Skills requires grasping three integration patterns that make the plugin powerful yet maintainable.

• Schema ingestion without dependencies: The SQL skill reads DDL, diagram descriptions, or plain-text table lists using regex heuristics—no external database drivers or parsing libraries required.
• Python script generation: The cohort skill outputs ready-to-run pandas scripts that can be copied into Jupyter notebooks for deeper exploration.
• Statistical transparency: The A/B test skill documents Z-score, pooled variance, and Bayesian formulas inline, though users receive plain-English summaries unless they request the math.
• Markdown-based extensibility: To add new SQL dialects, metrics, or visualizations, edit the corresponding skill file—no code recompilation or deployment needed.

Architecture: How Commands Orchestrate Skills

The PM Skills architecture separates concerns cleanly. Commands live in pm-data-analytics/commands/ and act as thin orchestration layers. Skills live in pm-data-analytics/skills/ and contain the full analytical logic in markdown. This pattern enables portability: the same skill folder works in Claude's /.claude/skills/, Gemini's /.gemini/skills/, or OpenCode's /.opencode/skills/ without modification.

The flow works as follows: the user invokes /write-query, the command handler parses arguments and loads the sql-queries skill, a structured prompt is sent to the AI assistant, and the skill's markdown logic guides the multi-step response. The command finalizes formatting and offers follow-up actions. This identical pattern applies to /analyze-cohorts and /analyze-test.

Summary

• PM Skills provides three core data analytics capabilities: SQL generation (/write-query), cohort analysis (/analyze-cohorts), and A/B test evaluation (/analyze-test).
• Each capability is implemented as a markdown skill file in pm-data-analytics/skills/, with command wrappers in pm-data-analytics/commands/ handling orchestration.
• The SQL skill supports BigQuery, PostgreSQL, MySQL, Snowflake, and other dialects through regex-based schema parsing.
• Cohort analysis outputs retention curves, heat-maps, and reproducible Python pandas scripts.
• A/B test analysis delivers statistical significance testing, confidence intervals, Bayesian probabilities, and actionable ship/extend/stop recommendations.
• The skill-first architecture requires zero code from product managers while remaining extensible through simple markdown editing.

Frequently Asked Questions

What AI assistants are compatible with PM Skills data analytics?

The PM Skills plugin works with Claude (Claude Code, Claude Cowork), Gemini, OpenCode CLI, and any AI assistant that loads skills from a /.{assistant}/skills/ directory. The skill files are standard markdown with no proprietary syntax, making them universally portable.

Do I need to know SQL or Python to use PM Skills for data analytics?

No. PM Skills is designed for zero-code analytics. The SQL and Python code is generated for you and explained in plain English. However, the skills do output runnable scripts if you want to customize or extend the analysis in a Jupyter notebook.

How does the SQL skill handle database schemas without a live connection?

The skill parses uploaded schema files using regex heuristics to extract table names, column names, keys, and relationships. It accepts DDL exports, diagram descriptions, or even plain-text lists. This approach eliminates the need for database drivers or live connections while still producing accurate, dialect-specific queries.

Can I modify the statistical methods used in A/B test analysis?

Yes. Because the skill logic lives in pm-data-analytics/skills/ab-test-analysis/SKILL.md, you can edit the markdown file to adjust significance thresholds, switch between frequentist and Bayesian approaches, or add alternative metrics. Changes take effect immediately without recompiling or redeploying the plugin.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}