Decision Path Modeling in E-commerce Sessions

Decision Framework: Path Structure × Commitment Strength
──────────────────────────────────────────

Browsing-only
(View / Category → Exit)
│
│ low intent
▼
Conversion ≈ 3%

──────────────────────────────────────────

Comparison-augmented
(Views ↔ Compare loops → Exit)
│
│ high hesitation
▼
Conversion ≈ 2%

──────────────────────────────────────────

Soft commitment
(Add-to-cart → extended interaction)
│
│ tentative commitment
▼
Conversion ≈ 2%

──────────────────────────────────────────

Hard commitment
(Checkout / Purchase → Exit)
│
│ decisive boundary
▼
Conversion ≈ 37%

This project explores how user decision-making behavior in e-commerce sessions can be better understood through decision path structure rather than raw activity metrics.

Using a semi-synthetic clickstream dataset, we demonstrate that how and when users commit matters far more than how much they interact.

---

Motivation

Traditional behavioral analytics often rely on scalar metrics such as:

• number of clicks
• session duration
• total page views

However, these metrics frequently fail to distinguish between:

• high engagement with low intent
• decisive behavior with minimal interaction

This project investigates whether decision path structure and commitment semantics provide stronger explanatory power for conversion outcomes.

---

Dataset

We use a semi-synthetic e-commerce clickstream dataset generated to reflect realistic user behavior patterns.

• Event-level data: timestamped user actions (view, compare, add-to-cart, checkout, purchase, exit)
• Session-level aggregation: derived behavioral features and path representations
• Path signature: compressed representation of action sequences within a session

The dataset is fully reproducible and designed to support structural analysis of decision processes.

---

Analytical Approach

The analysis proceeds in four stages:

1. Baseline EDA
   • Validate that session length and activity volume alone do not explain conversion.
2. Decision Path Structure
   • Abstract raw clickstreams into interpretable path patterns.
3. Commitment Semantics
   • Distinguish between soft commitment (add-to-cart) and hard commitment (checkout or purchase).
4. Decision Framework Construction
   • Combine path structure and commitment strength into a unified behavioral framework.

Throughout the analysis, emphasis is placed on interpretability and rule-based reasoning rather than black-box modeling.

---

Key Findings

1. Path length alone does not distinguish conversion outcomes.
   Converted and non-converted sessions exhibit nearly identical distributions of interaction volume.

2. Decision path structure provides strong separation of user intent.
   Commitment-touch paths show an order-of-magnitude higher conversion rate than browsing-only or comparison-augmented paths.

3. Comparison-heavy behavior is associated with decision friction.
   Sessions with extensive comparison actions display higher hesitation and lower conversion.

4. Soft commitment is a weak signal of purchase intent.
   Add-to-cart actions, regardless of timing, do not reliably predict conversion.

5. Hard commitment forms a decisive behavioral boundary.
   Checkout and purchase actions correspond to the highest conversion rates, shortest paths, and lowest hesitation.

---

Decision Path Framework

Based on these findings, we propose a qualitative decision framework:

• Browsing-only paths
  Passive exploration with minimal intent and near-zero conversion.

• Comparison-augmented paths
  High cognitive load and hesitation with limited payoff.

• Soft commitment paths
  Tentative intent that often leads to prolonged interaction without conversion.

• Hard commitment paths
  Decisive behavior characterized by efficient progression to purchase.

This framework highlights that commitment strength is more informative than commitment timing or activity volume.

Decision Framework (Visual Summary)

Segment	What it looks like	Conversion	Avg events	Avg hesitation
Browsing-only (no commitment)	View-heavy exploration → exit	2.9%	11.6	0.108
Comparison-augmented (no commitment)	Views + compare loops → exit	2.4%	10.4	0.238
Soft commitment (add_to_cart)	Add-to-cart present, often tentative	2.1%	10.8	0.142
Hard commitment (checkout / purchase)	Checkout/purchase boundary, decisive	37.2%	9.8	0.125

Takeaway:
Commitment strength is the dominant signal of conversion intent.
Soft commitment actions (add-to-cart) behave similarly to non-commitment paths, while hard commitment forms a clear behavioral boundary with dramatically higher conversion efficiency.

Detailed Analysis

The full exploratory analysis, including step-by-step reasoning, intermediate findings, and robustness checks, is documented in the following notebook:

• Exploratory Data Analysis & Decision Path Modeling
  notebooks/01_eda.ipynb

This notebook contains:

• Session-level EDA and validation
• Decision path structure abstraction
• Commitment semantics refinement
• Construction of the final decision framework

---

Implications

From a product analytics perspective, these results suggest:

• Early add-to-cart events should not be over-interpreted as strong intent signals.
• Friction-reduction efforts may be better targeted at comparison-heavy paths.
• Decision modeling should prioritize structural behavior patterns over aggregate metrics.

The framework is designed to be extensible and can serve as a foundation for downstream modeling or experimentation.

---

Repository Structure

behavior-state-modeling
├── data/          # Generated sample data
├── notebooks/     # EDA and analysis notebooks
├── docs/          # Project logs and documentation
├── src/           # Data generation and utilities
└── README.md

---

Notes

This project emphasizes interpretability and analytical reasoning rather than predictive optimization.
All conclusions are supported by reproducible analysis in the accompanying notebooks.