sboghossian Ops — PostHog Funnel Debugger
Purpose
A funnel analysis tells you where users are dropping off. A funnel debugger tells you why, and what to do about it. This skill wraps PostHog funnel analysis with a structured diagnostic methodology — segmenting drop-off by cohort, exploring user paths between steps, and generating concrete experiment hypotheses ranked by expected impact.
When to use this
Use this skill when:
- Conversion rate at a key funnel step has declined ≥10% week-over-week or month-over-month.
- A new feature was released and you want to understand its impact on the conversion funnel.
- You are preparing an experiment hypothesis and need data to confirm the problem exists.
- A stakeholder asks "where are users getting stuck?"
Canonical legal AI funnel
The default funnel for a legal AI product:
Step 1: Signup (account created)
Step 2: First substantive prompt sent
Step 3: First draft saved (or first document analysis completed)
Step 4: Matter created (user organizes work into a matter)
Step 5: Upgrade to paid plan
Adjust steps based on the specific product flow — this is a starting template.
Analysis steps
1. Run the funnel in PostHog
Configure the PostHog funnel with:
- Conversion window: 14 days (allow enough time for deliberate users)
- Counting method: unique users (not events)
- Date range: last 30 days (or align with the change you're investigating)
- Breakdown: by the relevant cohort dimension (persona, tier, acquisition channel)
2. Extract step-by-step conversion
For each step transition, record:
- Conversion rate (% of users who proceeded from step N to step N+1)
- Absolute number of users (don't let a high % mask a small absolute number)
- Median time from step N to step N+1
This produces a table like:
| Transition | Conversion | Median time to step |
|---|---|---|
| Signup → First prompt | 68% | 2 hours |
| First prompt → First draft saved | 41% | 1 day |
| First draft → Matter created | 52% | 3 days |
| Matter created → Upgrade | 18% | 12 days |
3. Segment drop-off
For the step with the worst conversion rate, compare cohorts:
- Lawyer vs consumer — does one persona drop off more?
- Acquisition channel — do users from LinkedIn convert differently than organic?
- Plan tier — do trial users behave differently than free direct signups?
- Time of signup — did a recent cohort perform worse (suggesting a product change broke something)?
4. Path analysis between steps
For users who dropped off at the worst step, use PostHog's path analysis to see what they did instead of proceeding:
- Did they navigate to the settings page? (confusion about the product)
- Did they trigger an error event? (bug preventing conversion)
- Did they come back the next day and convert? (delay, not abandonment)
- Did they exit the app immediately? (activation failure)
5. Generate three hypotheses
Based on the conversion data, segmentation, and path analysis, generate exactly three ranked hypotheses for the largest drop-off point:
Format each hypothesis as:
- Problem: What behaviour is causing the drop-off?
- Evidence: What data supports this?
- Experiment: What would we change to test this?
- Expected impact: How much could conversion improve?
Example:
Hypothesis 1 — Friction at the first draft step Problem: Users are typing long prompts but abandoning before saving a draft. Evidence: Median time at step 3 is 3 hours; path analysis shows 40% of dropoffs go to the error page. Experiment: Fix the PDF upload error affecting users who try to draft from an uploaded document. Expected impact: +8–12% conversion at step 3.
6. Recommend experiments
For the top hypothesis, link to [[ops-feature-flag-experiment-launcher]] with:
- A drafted hypothesis statement
- The primary metric (conversion rate at the drop-off step)
- Two guardrail metrics
- The cohort to test on
Output format
Deliver the funnel debug as a brief structured report:
- Funnel table (step-by-step conversion + median time)
- Worst-performing step identified
- Segmentation breakdown for that step
- Path analysis summary (top 3 paths for dropoffs)
- Three hypotheses (ranked by expected impact)
- Recommended experiment for the top hypothesis
Related skills
- [[ops-feature-flag-experiment-launcher]] — execute the experiment recommended by this analysis
- [[ops-posthog-cohort-builder]] — build the cohort breakdowns used in the funnel segmentation
- [[ops-churn-risk-detector]] — funnel dropoffs at the upgrade step feed the churn risk model