The questions we hear most often about what the dataset is, how it behaves, and how you can use it.
Completely. Every name, address, email, account, order, and financial record is fabricated. There is no personally identifiable information, no real customer or employee, and no real transaction anywhere in it. It is not derived from, sampled from, or anonymized from any real company's data — it's generated from nothing but configuration and models of business behavior.
A fictional U.S. wholesale distributor of industrial and office supplies. By default it does about $4.4 billion in revenue a year — roughly $12 million a day across ~50,000 orders — through 27 distribution centers, serving thousands of business customers across five commercial segments. Every one of those figures is a configurable default.
At the default scale, about 809 million rows across 62 interrelated tables — two full calendar years of daily activity. Reduced-scale editions preserve the same structure and relationships if you need a smaller footprint for quick iteration.
Parquet (columnar and compressed, ideal for data lakes and distributed compute), CSV (universally importable), and a fully documented PostgreSQL schema, ready to load. The data files are database-agnostic, so they load into any major engine unchanged.
Any of them. The schema (table definitions, keys, indexes, and constraints) can be provided in the dialect for your database — PostgreSQL, SQL Server, MySQL, Oracle, or Snowflake — with types and constraints translated to match each engine. The Parquet/CSV data itself loads into all of them without change.
The library of pre-built analytics views and materialized views that powers our dashboards is available as an optional add-on to the dataset. We provide them as regular views everywhere, or as true materialized views on the engines that support them, matched to your database.
Yes — that's the point. The whole business is produced from more than 3,300 individually adjustable values across roughly 30 domains: company size, footprint, segment mix, pricing and promotions, fill rates, return rates, wage bands, cost structure, and much more. We can deliver a custom configuration tuned to your scenario.
The PostgreSQL schema drives BI tools like Tableau, Power BI, Looker, and Metabase directly. The Parquet output slots into Spark, DuckDB, Snowflake, BigQuery, and Databricks. And of course it works with pandas, dbt, and anything that reads CSV or Parquet.
Every edition is validated before release against an automated suite of integrity and realism checks — referential integrity, arithmetic and financial identities (the margin waterfall reconciles, receivables tie to revenue, payroll and overhead reconcile), distribution properties, and business-rule constraints. An edition that fails is never shipped.
Yes. A given edition is fixed and versioned, so every recipient works from identical data. Results are comparable across teams, demonstrations repeat exactly, test suites stay deterministic, and a benchmark run today can be reproduced months from now.
Deliberately. Real businesses are messy, so the data includes a tail of unprofitable customers, delinquent accounts, late vendors, damaged returns, price exceptions, and cost outliers — so your analyses have something real to find rather than an artificially clean world.
You can freely share what you build with it — demos, dashboards, visualizations, screenshots, reports, and presentations can be shown publicly, embedded in courseware, and used in publications, with no confidentiality concerns (there's no real personal or financial data in it). The dataset itself, however, is licensed for your own use and is not for public redistribution — please don't republish, resell, or hand the raw data files to the public.
Request access using the button below. Tell us a little about what you're building and we'll set you up with the dataset and a walkthrough of the dashboards built on it.
