E-Commerce Conversion Funnel Analysis

Data-driven root cause analysis of 2.7M e-commerce events across 1.4M users to identify why 97.3% of viewers never add to cart — and what to do about it.

---

The Story

What

97.3% of viewers never add to cart. View→cart rate is 2.69% vs. a 5–10% industry benchmark. Cart→purchase is strong at 31%, meaning the problem is upstream — users are leaving before they engage, not after.

Where

The drop-off is not uniform. Conversion rates vary 30x across product categories. More critically, 60% of all views are on out-of-stock items, which convert at 1.7% vs. 4.5% for in-stock items. Availability alone is a direct, structural driver of the overall rate.

Who

Segmenting 1.4M users into four behavioral profiles reveals the aggregate funnel is four different behaviors averaged together:

Segment	Users	% of Total	Avg Items/Session	View→Cart	Cart→Purchase
Browser	1,368,715	97.2%	1.5	~0%	—
Cart Abandoner	27,146	1.9%	4.1	28%	0%
Decisive Buyer	7,920	0.6%	2.5	33%	100%
Researcher	3,799	0.3%	30.3	12%	71%

Browsers — 97% of users — average 1.5 items viewed per session and almost never add to cart. They are not failing to complete checkout; they are never engaging with the catalog deeply enough to find something they want.

Why

Three hypotheses, each grounded in the data:

1. Product discovery gap — Users viewing 15+ items per session convert at 44.7% vs. 1.4% for single-item sessions (33x gap). Most users never browse deeply enough to find relevant items.
2. Category-specific quality gap — Low-converting categories may have thinner content (fewer images, weaker descriptions, less social proof) relative to high-converting ones.
3. Consideration journey gap — 45.3% of purchasers needed more than one session before buying. There is no re-engagement mechanism to bring interested users back.

What to Do

See reports/root_cause_analysis.md for full experiment designs. The highest-leverage intervention is a recommendation carousel on product pages — it targets the Browser segment (97% of users) and is directly testable with a clean A/B design with ~18,000 users per arm and ~7–10 days runtime.

---

Key Metrics

Metric	Value	Industry Benchmark	Status
View → Cart	2.69%	5–10%	Below
Cart → Purchase	31.07%	25–35%	Good
Overall Conversion	0.83%	2–3%	Below
Cohort stability	±0.6pp week-over-week	—	Structural, not seasonal

---

Notebooks

Notebook	Question
01_data_exploration.ipynb	What does the raw data look like?
02_data_enrichment.ipynb	How do we add category and availability data?
03_funnel_analysis.ipynb	Where are users dropping off and what's the revenue opportunity?
04_category_analysis.ipynb	Is the drop-off uniform across categories?
05_session_analysis.ipynb	Does browsing depth predict conversion?
06_cohort_analysis.ipynb	Have rates changed over time?
07_user_segmentation.ipynb	Who are the users that never convert?

---

Reports

• reports/root_cause_analysis.md — Three root cause hypotheses with experiment designs and sample size calculations

---

Project Structure

funnel-analysis/
├── data/
│   ├── events.csv
│   ├── item_properties_part1.csv
│   ├── item_properties_part2.csv
│   ├── category_tree.csv
│   └── processed/
│       ├── events_enriched.csv
│       └── sessions.csv
├── notebooks/
│   ├── 01_data_exploration.ipynb
│   ├── 02_data_enrichment.ipynb
│   ├── 03_funnel_analysis.ipynb
│   ├── 04_category_analysis.ipynb
│   ├── 05_session_analysis.ipynb
│   ├── 06_cohort_analysis.ipynb
│   └── 07_user_segmentation.ipynb
├── reports/
│   └── root_cause_analysis.md
├── sql/
├── src/
└── requirements.txt

How to Run

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
jupyter notebook

Run notebooks in order starting from 01_data_exploration.ipynb.

---

Author: Matthew Yang