Investigate a revenue dip, onboarding drop-off and user segments with Claude Code, the PostHog MCP and your interview transcripts
A product person connects Claude Code to PostHog and a folder of company context and interview transcripts, then asks it to explain a revenue dip, a funnel drop and user segments.
Evidence: The author reports this. We have not checked it beyond reading the source.
The business problem
Quantitative events and qualitative interviews sit in different tools, so finding why a number moved takes hours of manual joining.
What was tried
Run Claude Code in a local folder connected to the PostHog MCP so it can query events with HogQL. In the folder, keep the company overview, ideal customer and goals, a data glossary, seasonality notes, a deployment log, and the interview and usability transcripts. Ask Claude whether a revenue dip is significant given seasonality and break it down by payment method. Ask it to map a funnel step to its event and search the transcripts for confusion. Have it propose user segments, then challenge its criteria, and test your own hypotheses about what drives retention before acting.
What was reported (not reported)
In the author's example Claude traced zero new invoice payments since Friday to a Thursday billing release, found a 33 percent drop at the AI setup step with a hypothesis about a local versus cloud choice, and showed mobile at 0.3 percent of sessions. No later fix or result is reported.
Limitations
One author's walkthrough of her own product with no data shown. HogQL results can differ from the PostHog interface because of sampling, timezone, default filters and caching, so check board-level numbers in the interface. The context folders and glossary need building and upkeep. Large transcript sets will strain the context window. The 'no mobile demand' conclusion may only reflect questions that were never asked. The author calls it a multiplier for product judgment, not a replacement.
What you need
A PostHog project with events, Claude Code, the PostHog MCP server, and written-up interview or usability transcripts. Time to build the context files.
Sources
- Else van der Berg (Substack) ↗ Firsthand write-up, published January 27, 2026
Source published: January 27, 2026. Last reviewed here: October 11, 2026. Spot a mistake? Tell us.
Tools in this workflow
- Claude Code: Runs the analysis against local context filesTry Claude Code
- PostHog: Provides the product event data through its MCP serverTry PostHog
Related workflows
Send a Monday business metrics email from Stripe, Notion and Sheets with Claude writing the narrative in n8n
A shared n8n template pulls last week's Stripe revenue, Notion deal pipeline and a Sheets row of support numbers, has Claude write a three-paragraph summary, and emails and posts it to Slack.
Find content gaps with a Claude Code agent and then cross-check against your live archive
A creator ran a published content gap agent with Perplexity and Firecrawl, got a scored gap report, then discovered her local files missed 37 published posts and the real problem was distribution.
Give Claude Code a website team of connectors: Search Console, DataForSEO, Screaming Frog and PostHog
A founder connects five MCP servers to Claude Code, each acting as an SEO analyst, market researcher, technical auditor or product analyst, with the prompts he uses, including a weekly rage-click report.
Get a Monday analytics report by having a Claude skill read scheduled Google Analytics and Search Console exports
A newsletter writer has Google Analytics and Search Console email six CSV reports each Monday, and a Claude skill compares the week to the last and saves a visual report and the raw data to a dated folder.