---
name: takumaoshiro/funnel-analysis-ユーザー行動ファネル分析
source: https://app.decimal.ai/s/takumaoshiro-funnel-analysis@1/SKILL.md
source_sha256: 94f49b1ac6bb
---

# Funnel Analysis (ユーザー行動ファネル分析)

> Analyze user behavior funnels to identify conversion bottlenecks and optimization opportunities — for product managers, UX designers, and growth teams.

---

## 1. Overview

- **Purpose**: Where are users dropping off in the conversion funnel, and what segments are most affected?
- **Audience**: Product managers, UX designers, growth teams
- **Frequency**: Weekly / Ad-hoc (after releases or campaign launches)
- **Time to complete**: 15–30 minutes per funnel

### When to use

Use this skill when conversion rates decline unexpectedly, after launching a new user flow, or when planning UX optimization sprints. It is especially valuable when you need segment-level visibility (mobile vs desktop, organic vs paid) to prioritize fixes.

---

## 2. Input Data

### Required

| Data Source | Format | Description |
|-------------|--------|-------------|
| GA4 event export | CSV / BigQuery | Pageviews, clicks, form_start, form_submit, purchase events with timestamps |
| User segments | CSV / API | Device type, traffic channel, new vs returning, geography |

### Optional

| Data Source | Format | Description |
|-------------|--------|-------------|
| Heatmap / session recordings | URL | Qualitative context for drop-off stages |
| A/B test assignments | CSV | Variant allocation per user for controlled comparison |
| CRM data | CSV / API | Post-conversion revenue, LTV for value-weighted funnel analysis |

### Data Quality Checks

- Verify date range covers at least one full business cycle (7 days minimum)
- Confirm event taxonomy is consistent (no duplicate or renamed events mid-period)
- Check for bot / spam traffic inflation (filter internal IPs, known bots)
- Ensure user_id or session_id is present for step-sequence validation

---

## 3. Analysis Steps

### Step 1: Define Funnel Stages
Map the intended user journey into discrete, measurable stages. Typical e-commerce example: Landing Page → Product View → Add to Cart → Checkout Start → Purchase. Confirm stage definitions with stakeholders before proceeding.

### Step 2: Calculate Stage-by-Stage Drop-off Rates
For each consecutive pair of stages, compute:
- **Absolute drop-off** = Users at Stage N − Users at Stage N+1
- **Drop-off rate** = (Users at Stage N − Users at Stage N+1) / Users at Stage N × 100%
- **Cumulative conversion** = Users at Stage N+1 / Users at Stage 1 × 100%

### Step 3: Segment Comparison
Break down each stage by key segments:
- **Device**: Mobile vs Desktop vs Tablet
- **Channel**: Organic vs Paid vs Direct vs Referral
- **User type**: New vs Returning
- **Cohort**: Weekly or monthly registration cohort

Flag segments where conversion at any stage is significantly below the overall average.

### Step 4: Time-Between-Steps Analysis
Measure median and P90 time between consecutive stages. Abnormally long times indicate friction or confusion. Compare across segments to identify device- or channel-specific delays.

### Step 5: Micro-Conversion Mapping
Identify intermediate actions (scroll depth, video play, tooltip hover, FAQ expand) that correlate with progression to the next stage. These are leading indicators for optimization experiments.

### Step 6: Synthesize and Prioritize
Combine drop-off severity, segment impact, and estimated traffic volume to produce a prioritized list of optimization opportunities. Use an impact × effort matrix for actionability.

---

## 4. Output Format

### Summary (top-level)
- Overall funnel conversion rate (top-to-bottom)
- Biggest single-stage drop-off (stage name, rate, absolute users lost)
- Top segment disparity (e.g., "Mobile CVR is 42% of Desktop CVR at Checkout")

### Detail sections
- Stage-by-stage table: Users, Drop-off %, Cumulative CVR
- Segment comparison matrix: rows = stages, columns = segments, cells = CVR
- Time-between-steps distribution per stage
- Micro-conversion correlation table

### Visualizations
- Horizontal funnel bar chart with drop-off percentages annotated
- Segment overlay line chart (one line per segment across stages)
- Heatmap of drop-off rates by stage × segment
- Time-to-next-step box plots per stage

---

## 5. Judgment Criteria

| Metric | Good | Warning | Critical |
|--------|------|---------|----------|
| Single-stage drop-off | < 30% | 30–50% | > 50% |
| Mobile CVR vs Desktop CVR | > 75% of desktop | 50–75% of desktop | < 50% of desktop |
| New user drop-off vs returning | < 1.5× returning | 1.5–2× returning | > 2× returning |
| Time-to-next-step (P90) | < 2× median | 2–5× median | > 5× median |
| Overall funnel CVR trend (WoW) | Improving or stable | Declining 1–2 weeks | Declining 3+ weeks |

### Decision Rules

- If drop-off exceeds 50% at any single stage, classify that stage as a **bottleneck** requiring immediate investigation.
- If Mobile CVR < 50% of Desktop CVR at the same stage, flag as a **mobile UX issue** — likely touch-target, layout, or load-speed related.
- If new-user drop-off is > 2× returning-user drop-off, flag as an **onboarding problem** — the first-time experience is failing.
- If time-between-steps P90 > 5× median, investigate page load performance and UI clarity at that transition.
- If a segment represents > 20% of total traffic and converts > 30% below average, it becomes a **high-impact optimization target**.

---

## 6. Examples

### Sample Input
```
date,user_id,device,channel,event,timestamp
2026-04-01,u001,mobile,organic,page_view,2026-04-01T10:00:00
2026-04-01,u001,mobile,organic,product_view,2026-04-01T10:01:12
2026-04-01,u001,mobile,organic,add_to_cart,2026-04-01T10:03:45
2026-04-01,u002,desktop,paid,page_view,2026-04-01T10:05:00
2026-04-01,u002,desktop,paid,product_view,2026-04-01T10:05:30
2026-04-01,u002,desktop,paid,add_to_cart,2026-04-01T10:06:10
2026-04-01,u002,desktop,paid,checkout_start,2026-04-01T10:07:00
2026-04-01,u002,desktop,paid,purchase,2026-04-01T10:09:22
```

### Sample Output
```
## Funnel Summary (2026-04-01 ~ 2026-04-07)

Overall Conversion: 3.2% (Page View → Purchase)
Total users entering funnel: 12,480

| Stage           | Users  | Drop-off | Cumulative CVR |
|-----------------|--------|----------|----------------|
| Page View       | 12,480 | —        | 100%           |
| Product View    |  8,120 | 34.9%    | 65.1%          |
| Add to Cart     |  2,840 | 65.0%    | 22.8%          |
| Checkout Start  |  1,210 | 57.4%    | 9.7%           |
| Purchase        |    400 | 66.9%    | 3.2%           |

### Bottleneck: Product View → Add to Cart (65.0% drop-off)
- Mobile drop-off: 72.1% vs Desktop: 54.3%
- New users: 71.8% vs Returning: 55.2%

### Priority Actions
1. [HIGH] Mobile product page UX — CVR 41% of desktop (threshold: <50%)
2. [HIGH] New user product discovery — 2.1× returning user drop-off
3. [MED] Checkout form simplification — 57.4% drop-off, P90 time 4.2min
```

---

## 7. Do's and Don'ts

### Do
- Always define funnel stages collaboratively with stakeholders before analysis
- Segment by at least device and channel — aggregate funnels hide critical issues
- Compare against a baseline period (previous week, previous month) for context
- Include absolute user counts alongside percentages to convey business impact
- Validate that funnel stages are sequential (users actually progress in order)

### Don't
- Don't report only overall conversion — segment-level insights drive action
- Don't assume linear funnels; check for skip-stage and loop-back patterns
- Don't ignore low-traffic segments — they may represent high-value users
- Don't conflate correlation with causation in micro-conversion analysis
- Don't present findings without actionable recommendations

---

## 8. Agent Guide

When executing this analysis skill:

```
You are performing Funnel Analysis (ユーザー行動ファネル分析).
Key rules:
- Calculate drop-off rates for EVERY consecutive stage pair, not just overall CVR.
- Always segment by device (mobile/desktop) and channel (organic/paid) at minimum.
- Flag any single-stage drop-off >50% as a bottleneck.
- Flag mobile CVR <50% of desktop CVR as a mobile UX issue.
- Flag new-user drop-off >2× returning-user drop-off as an onboarding problem.
- Include absolute user counts alongside percentages.
- Produce a prioritized action list ranked by impact × traffic volume.

Refer to the full SKILL.md for thresholds, output format, and detailed steps.
```