posthog-investigationlisted

Investigate what happened to a product's funnel over a period with PostHog, and name the one bottleneck with evidence. Reads the dashboards and goals the founder already built, rules out false alarms (incomplete period, normal variance), walks a list of causes (deploy, flag, traffic mix, tracking bug, new vs returning users, device, outage), finds who dropped and where they went instead, reads event timelines, session replays and heatmaps of representative visitors, compares before and after a deploy, cross-checks on a control segment, and writes a prioritized report with a confidence level. Use it whenever the user asks "why no conversions / no sales / no signups", "what happened in the last 24-72 h", "conversion dropped", "analyze my sessions / replays / dashboard / traffic", "is the new onboarding working", "did the deploy break something", or has just turned on ads, even if they never say "funnel". For one paid campaign judged by its UTMs, use posthog-paid-funnel; for an A/B test, posthog-experiment-reade
josephbedminster/indie-marketing-skills · ★ 1 · AI & Automation · score 74

Install: claude install-skill josephbedminster/indie-marketing-skills

# PostHog funnel investigation The user is usually a solo founder who shipped things, turned on ads and saw a number they don't like. They want a reconstruction, not a data dump: what changed, what the real volume is once the noise is removed, where people drop, why (with evidence), and what to do first. One bottleneck, named, with numbers and a confidence level. **Before anything:** read `.claude/posthog-funnel-map.md`. If it doesn't exist, run the `posthog-funnel-map` skill first. Never query an event name that is not in the map or confirmed with `read-data-schema`. Calibrate every judgment on the map's business profile (baseline sales per week, traffic, read window). Tool conventions (MCP tool names, `info` before first use, `search` when a name changed) are the same as in `posthog-funnel-map`. Every call in this skill is a read, except creating an annotation or a replay scanner, which needs the user's yes. **Project conventions** in the map (report language and style, report location, spend sources, extra context sources) override this skill's defaults. ## Step 0: what changed (in parallel, before any analytics) Numbers without context lead to wrong stories. Collect: 1. **Code shipped.** In the repo: `git log --since="<date>" --pretty='%h %ad %s' --date=iso` on the main branch, and `gh pr list --state merged --search "merged:>=<date>"` if GitHub is available. Keep the timestamp of each change that touches onboarding, pricing, paywall, signup or checkout: