Generate recurring weekly reports combining KPIs and pull requests
A product leader uses Claude Code for automatic recurring reports on what developers shipped and whether features helped or hurt KPIs. Task: David Kelly uses Claude Code to build a system that fetches weekly GitHub pull requests and Metabase KPI data. The workflow compares period-over-period KPI changes with PR content to score the likelihood of each PR affecting specific metrics. Zen MCP tools connect OpenAI and Gemini APIs to provide cross-model consensus for enhanced accuracy. Kelly reviews these automated reports to prioritize product iterations and validate feature impacts.; tools: Claude Code. Human role: Source evidence is unavailable regarding specific human roles in this workflow.; limitations: A paid Claude Pro plan ($20/month) is required; API usage is discouraged due to cost accumulation. The workflow depends on accurate KPI definitions in Metabase and consistent PR tagging in GitHub. Correlation does not imply causation, so AI scores may misattribute KPI changes to unrelated PRs. Human validation is necessary to interpret confidence scores and avoid acting on false positives..
Tools
Published on Agent Work Atlas:
Overview
A product leader uses Claude Code for automatic recurring reports on what developers shipped and whether features helped or hurt KPIs.
Workflow
David Kelly uses Claude Code to build a system that fetches weekly GitHub pull requests and Metabase KPI data. The workflow compares period-over-period KPI changes with PR content to score the likelihood of each PR affecting specific metrics. Zen MCP tools connect OpenAI and Gemini APIs to provide cross-model consensus for enhanced accuracy. Kelly reviews these automated reports to prioritize product iterations and validate feature impacts.
Reported result
The author receives weekly reports indicating whether shipped features helped or hurt KPIs, which he uses to guide product iteration. No specific accuracy figures or correlation success rates are provided in the source. The system relies on AI-generated confidence scores rather than proven causal links. Effectiveness is described qualitatively as helping him 'iterate accordingly' without quantified business impact metrics.
Human role
Source evidence is unavailable regarding specific human roles in this workflow.
Limitations
A paid Claude Pro plan ($20/month) is required; API usage is discouraged due to cost accumulation. The workflow depends on accurate KPI definitions in Metabase and consistent PR tagging in GitHub. Correlation does not imply causation, so AI scores may misattribute KPI changes to unrelated PRs. Human validation is necessary to interpret confidence scores and avoid acting on false positives.
Try it
Act as a product analytics engineer. I need you to build a Python script using Claude Code that generates a weekly KPI-PR correlation report. Inputs: 1. Metabase API credentials and dashboard IDs for [replaceable: KPI metrics like signups, conversion]. 2. GitHub repository names and access tokens for [replaceable: product repos]. 3. Date range: last 7 days. Steps: 1. Fetch weekly PRs (title, description, changed files) from GitHub. 2. Fetch current and previous week KPI values from Metabase. 3. Use Claude Code to analyze PR content and score correlation likelihood (0-1) with KPI changes, leveraging Zen MCP for multi-model consensus if available. 4. Generate a Markdown report listing top PRs with high confidence scores and their associated KPI impacts. Safeguards: - Run a dry test with 3 sample PRs first. - Do not auto-publish; save output to [replaceable: local path] for my review. - Flag any PRs with low confidence scores (<0.3) as 'Uncertain'. - Ask me to confirm the KPI mapping logic before full execution.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.