Activation Funnel Expert

Overview

A funnel is the single most useful diagnostic tool a growth or onboarding PM owns. It turns a fuzzy product story ("users drop off somewhere") into a numbered, actionable picture ("76% land, 41% start setup, 9% finish setup, 4% take the activation action -- the biggest drop is between start and finish setup at 32 percentage points").

This skill specifies funnel structures using Dave McClure's AARRR (Acquisition, Activation, Retention, Revenue, Referral) and its broader cousin AAARRR (which adds Awareness on the front), and analyzes them with a stdlib Python tool (funnel_analyzer.py). The tool ingests a JSON funnel definition (stages with counts) and outputs stage-by-stage conversion and drop-off, a Mermaid flowchart, and a bottleneck call-out — in all six SHAREDOUTPUTSCHEMA formats so the analysis travels into Jira, Linear, Confluence, Notion, or a PR.

The activation step is the centerpiece. Sean Ellis defined the "activated user" as one who has done the thing that statistically predicts retention (Slack's 2000 messages, Facebook's "7 friends in 10 days", Dropbox's "1 file in 1 folder on 1 device"). Pin the activation event before you optimize the funnel that leads to it.

Core Capabilities

• Funnel structure — define stages as events using AARRR or AAARRR; compress to 4-7 stages.
• Conversion + drop-off math — per-stage conversion, absolute and relative drop, cumulative conversion, bottleneck detection (largest absolute vs largest relative drop).
• Activation event definition — Sean Ellis framework to pin the predictive "aha" event (count + window + action).
• Counter-metric pairing — guard every stage against gaming; leading-vs-lagging indicator design.
• Six-format output — render the analysis + Mermaid diagram for Jira, Linear, Confluence, Notion, or a PR.

When to Use

• New onboarding design -- specify the funnel events from signup to value.
• Drop-off diagnosis -- signups grow but completion is flat; where is the leak?
• Activation metric definition -- pick the "aha" event that predicts long-term retention.
• Cross-functional alignment -- get marketing, product, and growth to agree on one funnel definition.
• QBR / A/B test design -- one canonical funnel drives the conversation and makes primary + counter-metrics obvious.

When NOT to use: pre-PMF discovery (use discovery/); top-of-funnel channel attribution (marketing tools); revenue-cohort retention (data-analytics domain); when events are not instrumented.

Clarify First

Before analyzing the funnel, settle the inputs below. When one is missing or vague and would change the result, ask about it; when it wouldn't, use a sensible default and say which one you used.

• Stages + counts: the actual events and the user count at each step (drives every conversion/drop-off number and the bottleneck call-out).
• Activation event: the Sean Ellis "aha" event that defines an activated user (sets the funnel's centerpiece stage and the activation rate).
• Cohort window: the time period the counts are drawn from (snapshot funnels mix cohorts and lie; this scopes every number).

Ask about the two or three inputs that matter most for this request, not the whole list. If the user wants a result straight away, produce it and state your assumptions at the top.

Workflow

1. Settle the Clarify First inputs; pin the activation event (count + window + action) with assets/activation_metric_worksheet.md before touching the funnel.
2. Define 4-7 stages as events for one cohort in the shape of assets/sample_funnel.json, each with a counter-metric.
3. Run funnel_analyzer.py. It reports the largest absolute drop, the worst conversion rate, and the worst conversion past acquisition; start with the last one, because top-of-funnel stages nearly always lose the most and say the least.
4. Done when the chosen bottleneck has a hypothesis, a primary metric and a counter-metric, and the activation event is shown to predict retention.

python3 project-management/execution/activation-funnel/scripts/funnel_analyzer.py --input project-management/execution/activation-funnel/assets/sample_funnel.json --format markdown
python3 project-management/execution/activation-funnel/scripts/funnel_analyzer.py --demo --format mermaid

Formats: markdown, json, mermaid, confluence, notion, linear (the past-acquisition bottleneck appears in markdown and json). Exit 0 on success, 2 on bad input (missing file, invalid JSON, malformed stages).

Anti-Patterns

Activation event that does not predict retention

Mistake: Activation is set to "clicked Get Started on day 1" without checking whether that event predicts day-30 retention.
Why it happens: An easy-to-instrument event is at hand, and validating it needs a cohort analysis nobody has run.
Instead: Compare 30-day retention for users who hit the candidate event with users who did not, and adopt it only when the lift is large. Use assets/activation_metric_worksheet.md and re-validate every 1-2 quarters.

Local stage win, funnel loss

Mistake: Removing six signup fields lifts signup conversion 40% and the change ships. Two weeks later activation is down 20%.
Why it happens: Each experiment is judged on the stage it changed.
Instead: When you optimize a stage, measure the next one or two stages downstream and report net activated users. Pair each stage metric with a counter-metric before the test starts.

Biggest drop treated as biggest opportunity

Mistake: The team works on a 40pp signup-to-activation drop while a 25pp drop at a later, higher-value stage is worth several times more revenue.
Why it happens: The funnel_analyzer.py bottleneck call-out ranks by drop size, and that is the easiest number to quote.
Instead: Treat the bottleneck call-out as a starting point. Weight each drop by what a user at that stage is worth (LTV or revenue) before choosing where to invest.

Nine more, each with a bad and a good example: references/red-flags.md.

References

Load the reference that matches the task — keep this file lean and pull detail on demand:

• references/funnel-analysis-playbook.md — full AARRR/AAARRR stage tables, the aha-event definition method, conversion/drop-off math, counter-metric pairing, leading-vs-lagging indicators, the step-by-step workflow, the funnel_analyzer.py reference (flags, input JSON, Mermaid sample), troubleshooting, and success criteria. Read when building or analyzing a funnel.
• references/pirate-metrics-deep-dive.md — McClure's AARRR + AAARRR, Andrew Chen funnel mechanics, Balfour's Four Fits, growth loops, and NSM-funnel linkage. Read when designing the overall growth model around the funnel.
• references/activation-aha-moment-patterns.md — Ellis's framework plus 12 worked activation-event examples (Slack, FB, Dropbox, Airbnb, Notion, Spotify, Duolingo, Twitter, HubSpot, Pinterest, LinkedIn, Figma). Read when choosing or validating the activation event.
• references/red-flags.md — concrete examples of how funnel output goes wrong, why it's bad, and how to fix it. Read when reviewing a funnel analysis or diagnosing a misleading chart.
• assets/funneldesigncanvas.md — workshop canvas for defining the funnel. Use in a design session.
• assets/activationmetricworksheet.md — worksheet for picking the activation event. Use when pinning the aha event.
• assets/sample_funnel.json — a working JSON example matching the tool input. Use as a starting template.

Scope & Limitations

In Scope: funnel definition (events, stages, transitions) via AARRR/AAARRR; conversion + drop-off math and bottleneck detection; Mermaid rendering; activation-event selection (Ellis); counter-metric pairing; cohort-vs-snapshot distinction; all 6 output formats.

Out of Scope: pulling raw event data (input is JSON — use Amplitude/Mixpanel/PostHog/Looker); statistical significance testing (discovery/brainstorm-experiments/, data-analytics); cohort-retention analysis; channel attribution (marketing tools); building the UX/A/B test; forecasting.

Caveats: a funnel implies a linear flow — real products branch and re-enter, so complement with branched analysis. Snapshot funnels mix cohorts and lie; always specify the cohort window. Top-stage gains amplify downstream and can mislead investment — look at both absolute and relative drop. Re-validate the activation event against fresh retention cohorts every 1-2 quarters.

Files