ziadatalabs/FreeSyntheticEcommerceOrders50M

Free Synthetic E-Commerce Orders (50M)

A free, fully synthetic dataset of 50,000,000 e-commerce order records, built for developers, analysts, and data scientists who need realistic retail order data for analytics pipelines, recommendation prototypes, BI dashboards, or return-prediction models.

Every value in this dataset is artificially generated. No real customers, no real orders, no real PII. The relationships between fields are deliberately consistent — order totals are computed from real line math, shipping follows a real threshold rule, and return rates vary by product category the way they do in the real world — so a model trained or tested against this data learns genuine structure rather than noise.

Schema

Column	Type	Description
order_id	string	Unique identifier for the order line
customer_id	string	Synthetic customer reference (repeats — customers place multiple orders)
order_date	string	ISO date (YYYY-MM-DD)
product_category	string	One of: electronics, clothing, home, beauty, sports, books, toys, grocery
product_id	string	Synthetic SKU (category-prefixed)
quantity	int	Number of units in the line
unit_price	float	Per-unit price (USD), within category-appropriate ranges
order_total	float	quantity × unit_price + shipping_cost
payment_method	string	One of: credit_card, debit_card, paypal, gift_card, bnpl
shipping_cost	float	Shipping charge (USD) — free above the order threshold
order_status	string	One of: delivered, shipped, processing, returned, cancelled
discount_applied	bool	Whether a discount/promo was applied

How the correlations work

• order_total is genuinely computed as quantity × unit_price + shipping_cost on every row — not a random number.
• shipping_cost is free for orders above the free-shipping threshold, and a tiered charge below it.
• Return rates vary by category — clothing and electronics are returned far more often than books or grocery, matching real retail patterns.
• unit_price ranges are category-appropriate (grocery items are cheap; electronics span a wide, higher range).

Format

Single Parquet file, Snappy compression, ~1.36 GB, 50,000,000 rows.

Quick start

import pandas as pd
df = pd.read_parquet("ecommerce_50M.parquet")

from datasets import load_dataset
ds = load_dataset("ziadatalabs/FreeSyntheticEcommerceOrders50M")

import duckdb
con = duckdb.connect()
con.sql("SELECT * FROM 'ecommerce_50M.parquet' LIMIT 10").show()

Notes

• All customers, orders, and SKUs are fabricated — no real individuals or businesses represented.
• order_total already includes shipping_cost — don't double-count shipping.
• Use order_status (returned vs delivered) as a return-prediction target; product_category is the primary signal.
• customer_id repeats across orders, so the data supports per-customer aggregation and repeat-purchase analysis.

License & Usage

Released under CC BY-NC 4.0 — free for personal, research, and educational use, with attribution. Not licensed for commercial use.

No real people, orders, or businesses are represented in this data. It is entirely synthetic.

---

Created by Zia Data Labs. Questions or feedback: zia.data.team@protonmail.com