BeautyCommerceOS is a large-scale synthetic enterprise data warehouse designed to benchmark autonomous AI agents and analytics systems operating in realistic business environments.

It simulates the full lifecycle of a modern global beauty ecommerce company — from user behavior to financial reconciliation — including the ambiguity, inconsistency, and cross-functional complexity found in real enterprises.

🧠 Why this dataset exists

Most datasets teach analysis on clean tables.

BeautyCommerceOS teaches something harder:

Cross-domain reasoning across a messy, real-world enterprise.

It is designed to evaluate whether AI systems can reason across:

conflicting KPIs across departments
delayed revenue recognition
attribution uncertainty across marketing channels
inventory and supply chain mismatches
finance vs marketing reporting divergence
experimentation interference effects

This makes it suitable for evaluating LLM agents, autonomous analysts, and decision intelligence systems.

🏗️ Dataset Structure

The dataset follows a modern medallion warehouse architecture:

🟤 Bronze Layer (Raw Behavioral Data)

sessions
clickstream events
anonymous users

⚪ Silver Layer (Conformed Dimensions)

products
SKUs
brands
suppliers
warehouses
identity mapping

🟡 Gold Layer (Business & Financial Truth)

orders
order_items
payments
refunds
campaigns
attribution
inventory
shipments
invoices
profit & loss (P&L)

📊 Key Capabilities

BeautyCommerceOS supports evaluation of:

🧠 AI Agent Reasoning

multi-step business question answering
cross-table joins across domains
causal inference under noisy signals

📈 Marketing Intelligence

attribution modeling
ROAS vs profit divergence
influencer impact analysis
channel cannibalization effects

🚚 Supply Chain Analytics

stockout impact analysis
fulfillment delay tracking
warehouse performance comparison

💰 Financial Reconciliation

revenue recognition delays
finance vs marketing mismatches
margin decomposition

🧪 Experimentation Analysis

A/B test evaluation
treatment contamination
causal uplift estimation

⚠️ Realism & Complexity

Unlike traditional synthetic datasets, BeautyCommerceOS intentionally includes:

inconsistent attribution signals
delayed financial reconciliation
missing or noisy event data
KPI definition conflicts across teams
operational distortions across systems

These are included to reflect real enterprise environments and enable robust evaluation of reasoning systems.

🔍 Example Evaluation Tasks

BeautyCommerceOS can be used to evaluate systems on questions such as:

📈 Business Performance

Why did revenue increase while profit declined?
Which channels generate the lowest long-term customer value?

🧠 Attribution & Marketing

Which attribution model best explains observed revenue?
Are influencer campaigns profitable after refunds and returns?

🚚 Operations

Which warehouses contribute most to fulfillment delays?
How do stockouts impact downstream revenue loss?

💰 Finance

Why do finance and marketing report different revenue figures?
What is the true margin after operational adjustments?

🤖 AI Agent Benchmarking

Can an autonomous agent reconcile conflicting KPIs across systems?
Can it identify root causes across marketing, finance, and logistics?

🚀 Intended Use

This dataset is designed for:

autonomous agent evaluation
LLM reasoning benchmarks
analytics engineering practice
causal inference research
BI system testing
data warehouse simulation
decision intelligence systems

🚫 Not Intended For

real-world financial forecasting
production decision-making
regulatory reporting
personal or sensitive data usage

🔒 Synthetic Data Statement

All data in this repository is fully synthetic and generated programmatically.

No real users, transactions, or personally identifiable information are included.

📦 Format

Columnar storage: Parquet
Architecture: Medallion (Bronze / Silver / Gold)
Structure: Partitioned data warehouse
Size: < 10 GB total

📜 License

Licensed under CC BY 4.0.

You are free to:

use
modify
redistribute
build upon

with attribution.

🏁 Vision

BeautyCommerceOS bridges the gap between:

clean academic datasets
and real-world enterprise complexity

It is designed to test whether modern AI systems can move beyond simple data analysis into true enterprise-level reasoning under ambiguity.

Downloads last month: 439

Total file size:

2.76 GB