Purpose
Turn raw funnel drop-off data into a ranked, hypothesis-driven experiment backlog that moves users from signup to first value faster. Pinpoints the highest-leverage stage to fix and produces specific UX/onboarding changes — each framed as a falsifiable hypothesis with a predicted mechanism and a success metric — so a growth team can stop guessing and start shipping experiments that compound.
When to Use
Invoke this skill when the user asks to:
- Improve activation rate or reduce onboarding drop-off
- Diagnose where new users stall before reaching first value (the "aha moment")
- Generate UX or onboarding experiments to fix a leaky funnel stage
- Build a prioritized activation experiment backlog
- Figure out why signups aren't converting into active users
- Sequence a 30-day test plan for the activation funnel
Inputs Required
Before starting, collect:
- Funnel stages: The ordered steps from signup to activation (e.g., signup → email verify → setup wizard → first project created → invite teammate)
- Stage conversion data: Conversion or drop-off rate between each stage (percentages or raw counts)
- Activation metric: The defined "activated" event and current activation rate (e.g., "created first dashboard within 7 days — 23%")
- Product type and core value: What the product does and the moment a user first feels value
- User segment: Primary persona or use case being optimized (self-serve, sales-assist, role, etc.)
- Known friction signals: Support tickets, session recordings, survey quotes, or qualitative notes if available
- Constraints: Engineering bandwidth, no-code/experimentation tooling available, anything off-limits
If stage-by-stage data is missing, ask for the single biggest known drop-off point and proceed; if qualitative signals are absent, note which hypotheses are lower-confidence as a result.
Instructions
Step 1: Map the Funnel and Quantify the Leak
Lay out the funnel stage by stage with conversion between each step.
- Calculate the drop-off magnitude at each stage (percentage of users lost).
- Calculate the recovery value of each stage: a stage where 60% of users drop is worth more than one where 10% drop, weighted by how many users reach it.
- Identify the highest-leverage stage = largest absolute number of users lost, not just the highest percentage. Losing 50% of 10,000 users beats losing 80% of 500.
- Flag any stage where conversion is below typical PLG benchmarks: signup → activation often lands 20–40%; setup-completion stages should clear 60%+; email verification below 70% signals a deliverability or motivation problem.
Step 2: Diagnose the Drop-Off Type at Each Leaky Stage
For each stage losing meaningful volume, classify the friction into one of five types — this determines the fix:
- Motivation gap: User doesn't yet see why the next step is worth it. Fix with value reinforcement, social proof, progress framing.
- Ability friction: The step is too hard, too long, or unclear. Fix with simplification, defaults, progressive disclosure, removing fields.
- Trigger weakness: User intends to continue but isn't prompted at the right moment. Fix with in-app nudges, email/lifecycle triggers, empty-state CTAs.
- Expectation mismatch: What the user expected differs from what they got. Fix with clearer setup framing, demo data, or repositioning the step.
- Time-to-value distance: First value is too many steps away. Fix by shortening the path, pre-populating data, or delivering a partial "aha" earlier.
State which type dominates at each leaky stage and the evidence (data shape + any qualitative signal) supporting that diagnosis.
Step 3: Generate Hypothesis-Driven Improvements
For each leaky stage, produce 3 distinct experiment hypotheses. Each hypothesis must follow this exact structure:
- Hypothesis: "We believe that [specific change] for [user segment] will [expected behavioral outcome] because [mechanism rooted in the diagnosed friction type]."
- The change: Concrete UX/onboarding modification — described specifically enough to hand to a designer (what screen, what element, what copy direction, what flow change).
- Why it should work: The behavioral or product reasoning tied to the Step 2 diagnosis. No generic "best practice" claims.
- Predicted lift: Directional estimate on the stage's conversion (e.g., "+5–8 pts on setup completion") with confidence: High / Medium / Low.
- Effort: T-shirt size (S / M / L) reflecting eng + design cost.
- Success metric: The single primary metric that proves or kills the hypothesis, plus a guardrail metric to watch for unintended downstream damage.
Generate variety across hypotheses — at least one low-effort/high-leverage option per stage, and avoid proposing three variations of the same idea.
Step 4: Rank the Experiment Backlog
Score every hypothesis on an ICE-style frame: Impact (predicted lift × users affected), Confidence (evidence strength + diagnosis clarity), Effort (inverse of T-shirt size). Produce a single ranked backlog table across all stages, top to bottom. Lead with the changes that are cheap, high-confidence, and touch the highest-volume leak.
Step 5: Sequence a 4-Week Test Plan
Convert the top of the backlog into a runnable sequence:
- Week 1: Highest-ranked, lowest-effort experiment on the biggest leak — what changes, what stays constant, sample-size note.
- Weeks 2–4: Subsequent experiments, ordered so learnings compound and you avoid running conflicting tests on the same stage simultaneously.
- For each: how to read a winner (primary metric movement + significance), and a stop/iterate rule if the test underperforms.
- Note any tests that should run sequentially rather than in parallel because they affect the same stage.
Output Format
- Structure: Funnel map → per-stage diagnosis → hypotheses grouped by stage → ranked backlog table → 4-week test sequence
- Length: 1,000–1,400 words
- Format: Structured document — funnel as a stage list with drop-off figures, hypotheses as labeled blocks, backlog as a ranked table, test plan as a week-by-week list
- Sections: Funnel & Leak Analysis → Drop-Off Diagnosis → Hypotheses by Stage → Ranked Experiment Backlog → 4-Week Test Sequence
- Tone: Growth specialist — evidence-led, opinionated about priority, allergic to vague "best practice" advice
Quality Checklist
Before delivering output, verify:
- [ ] The highest-leverage stage is identified by absolute users lost, not just highest drop-off percentage
- [ ] Each leaky stage has a single dominant friction type assigned from the five categories
- [ ] Every hypothesis uses the full "We believe... will... because..." structure with all six fields filled
- [ ] Each hypothesis has a single primary success metric AND a guardrail metric
- [ ] The backlog is ranked top-to-bottom with at least one low-effort/high-leverage option surfaced near the top
- [ ] The 4-week plan flags any tests that must run sequentially to avoid contaminating the same stage
Notes
- The biggest activation wins usually come from shortening time-to-value, not adding more onboarding — be suspicious of any hypothesis that adds steps rather than removing them.
- A high drop-off at email verification is almost always deliverability or motivation, not UX — check spam placement before designing screen changes.
- When qualitative signals (session recordings, support tickets) contradict the data shape, trust the qualitative for diagnosis and the quantitative for prioritization.
- Resist optimizing a late-funnel stage that few users reach; fixing the first major leak compounds value through every downstream step.
How to use this skill
This file is a Claude Code skill in Anthropic's standard format — markdown with YAML frontmatter at the top.
Installing in Claude Code:
~/.claude/skills/<folder-name>/SKILL.md— auto-loads in every session anywhere on your machine (recommended for skills you'll use across multiple projects)<your-project>/.claude/skills/<folder-name>/SKILL.md— auto-loads only when working in that project (recommended for project-specific skills)
Restart Claude Code or open a new session to register.
Using in Claude.ai, ChatGPT, Gemini, or other AI tools:
- Drag this file into a new chat, or copy and paste its contents
- Send your task or request as your next message — the AI will apply the skill where it fits
For Claude Code skill details: https://docs.anthropic.com/en/docs/claude-code/skills