Maycee Retail: A Synthetic Dataset for Realistic BI

Many retail demos are too small or too flat to answer serious analytical questions. Maycee Retail is SData's synthetic Australian retail dataset for practitioners who need realistic volume, connected business domains, and repeatable analysis without exposing customer or production data. It is designed for BI development, data engineering, analytics engineering, training, and AI evaluation.

Maycee Retail provides 10 years unified (3 free + 7 premium). The free tier covers 2017 to 2019 for accessible exploration, while licensed releases extend from 2020 through the available premium snapshot. This case study uses a licensed snapshot covering 1 January 2017 through 6 September 2026.

The model connects transactions and line items with customers, stores, regions, products, categories, promotions, and returns. That structure gives teams enough context to examine daily and monthly movement, seasonality, geographic differences, category mix, basket value, margin erosion, promotional behaviour, and return patterns instead of working from isolated totals.

BI and data teams can use Maycee Retail to build executive dashboards, dimensional models, DuckDB or warehouse transformations, data quality controls, SQL exercises, and portfolio projects. AI teams can use the same connected schema to prototype text-to-SQL, analytics agents, retrieval workflows, and evaluation scenarios against synthetic rather than sensitive enterprise data.

The dashboard below demonstrates that analytical range. Its licensed cache contains 836,560 transactions, 2,122,571 line items, and 47,287 returns. The views combine gross margin, return value rate, average order value, discount rate, month-over-month and year-over-year movement, regional contribution, category profitability, promotion impact, and category-level return behaviour. Shared business filters let users compare those signals across years, regions, categories, store types, and payment methods.

Maycee is also designed to fit practical local analytics workflows. Snapshot-scoped AWS credentials and the entitled S3 prefix are kept in an ignored environment file. A boto3 fetch step downloads only the required Parquet objects into an ignored cache, DuckDB performs local analytical preparation, and Pandas, Plotly, and Streamlit present the results. Credentials, licence records, and source Parquet files never appear in the dashboard or screenshots.

On the first measured development-machine run, 10,606 required objects totalling 284.9 MiB downloaded in 178.8 seconds. DuckDB prepared the dashboard tables in 72.648 seconds, giving an empty-cache-to-ready time of approximately 251.4 seconds, or 4 minutes 11 seconds. These measurements demonstrate one practical workflow and are not a performance guarantee.

The five screenshots show executive, sales, region and store, product and margin, and returns analysis produced from the licensed snapshot. The open dashboard repository provides a working reference implementation; Maycee Retail provides the substantial, connected dataset behind it. Explore the dataset tiers for your own BI, engineering, training, or AI evaluation work.