Dataset of the 'Contoso' Company

A collection of synthetic datasets simulating sales transactions from a fictional company. The dataset includes various related tables that contain essential business and operational data, useful for analyzing sales performance and other business insights. Key tables included in the package are: - "sales": Contains data on individual sales transactions, including order details, pricing, quantities, and customer information. - "customer": Stores customer-specific details such as demographics, geographic location, occupation, and birthday. - "store": Provides information about stores, including location, size, status, and operational dates. - "orders": Contains details about customer orders, including order and delivery dates, store, and customer data. - "product": Contains data on products, including attributes such as product name, category, price, cost, and weight. - "calendar": A time-based table that includes date-related attributes like year, month, quarter, day, and working day indicators. The package also includes 1,127 fictional internal business reports (weekly sales flashes, monthly reviews, regional reviews, pricing committee minutes, annual operating plans and industry notes) written about the trading from 2016 to 2025 in the larger cloud dataset, with a glossary of the terms they use, for building and testing document search and financial planning and analysis workflows with large language models. This dataset is ideal for practicing data analysis, performing time-series analysis, creating reports, or simulating business intelligence scenarios.

Contoso is a synthetic dataset containing sample sales transaction data for the fictional “Contoso” company. It includes various supporting tables for business intelligence, such as customer, store, product, and currency exchange data.

This dataset is perfect for practicing time series analysis, joins, financial modeling, or any business intelligence-related tasks.

It comes with a built-in dataset as well as the ability to create an in-memory database with duckdb

The package comes with the following tables:

• sales:
  • Contains information about sales transactions, including the total sales amount, customer, store, and product involved.
• customer:
  • Contains details about customers, such as customer key, name, address, and demographic information.
• store:
  • Contains information about stores, including store key, name, location, and related details.
• product:
  • Contains information about products, such as product key, name, category, and price.
• fx:
  • Contains foreign exchange rate data, mapping currency pairs to their exchange rates on specific dates.
• calendar:
  • Contains date-related information, including date, week, month, quarter, and year for use in time-based analysis.
• orders:
  • Contains information about individual orders, including order key, customer key, order date, and store information.
• orderrows:
  • Contains detailed line items for each order, including product key, quantity, and price for each item in the order.

Built into the package is the smallest version of the dataset: 7,794 sales lines across 3,242 orders and 3,165 customers.

Every column is documented in a machine-readable data-dict dictionary, shipped with the package. There is one per dataset, each in its own directory: extdata/small/ describes the bundled tables (and size = "small"), extdata/cloud/ describes the medium, large and mega sizes.

system.file("extdata", "small", "data-dict.yaml", package = "contoso")

Both validate against the spec, so the data-dict CLI can check either against its data, or render it as a browsable page. Point it at the directory:

Using labelled::look_for(), you can also see the columns’ labels from the labelled package.

Inspiration to using labelled comes from Crystal Lewis excellent blog post

For larger datasets, use create_contoso_duckdb() with one of the following sizes:

Column names and types are identical across all four sizes; only the row counts differ.

Data Storage

The larger datasets are stored as Parquet files on Cloudflare R2 cloud storage and streamed directly into DuckDB via the public URL:

https://pub-6aa63519a4b945948cb8c88949b320ca.r2.dev

Source

The data is originally sourced from the sqlbi github site

Dataset overview

The relationship keys that join each of the tables are listed below.

Installation

You can install the package from CRAN:

install.packages("contoso")

Or install the development version from Codeberg:

# install.packages("remotes")
remotes::install_git("https://codeberg.org/usrbinr/contoso")

Example

library(contoso)

# Create a DuckDB connection to Contoso datasets
db <- create_contoso_duckdb(size = "medium")

# Access the sales dataset
db$sales |> head()

# Launch the DuckDB UI to explore all tables interactively
launch_ui(db$con)

# Clean up when done
DBI::dbDisconnect(db$con, shutdown = TRUE)

Business reports

The package also ships 1,127 fictional internal business reports about Contoso’s trading from May 2016 to the end of 2025: weekly sales flashes, monthly e-commerce, supply chain and plant reviews, quarterly regional reviews, pricing committee minutes, operations notices, industry notes and each year’s Annual Operating Plan, with a glossary of the terms they use. Their figures come from the “mega” dataset, and a few contain deliberate errors, for testing document search and FP&A assistants.

reports <- read_reports() # one row per report, Markdown text in `body`
export_reports("contoso-reports") # or write them as one YAML file each

See the “Business reports” and “Searching the reports with ragnar” articles on the package website.