Skip to content

0003 — Use the Olist e-commerce dataset

  • Status: Accepted
  • Date: 2026-07-22

Context

We needed a dataset that is (a) freely and legally usable in a public portfolio, (b) genuinely business-shaped — products, orders, customers — and (c) rich enough to practise real data modelling, not a pre-cleaned toy table.

Candidates considered: Microsoft Contoso and AdventureWorks (clean, BI-native, but pre-modelled) and the Olist Brazilian e-commerce dataset (real, transactional, spread across multiple files).

Decision

Use the Brazilian E-Commerce Public Dataset by Olist (Kaggle: olistbr/brazilian-ecommerce).

Consequences

  • ✅ Real-world data (~100k orders, 2016–2018) with customers, products, sellers, payments, reviews and geolocation — an authentic portfolio story.
  • ✅ Delivered as nine related CSVs, so we must design the star schema ourselves — the most valuable modelling practice.
  • ⚠️ Licensed CC BY-NC-SA 4.0: non-commercial use only, attribution required, derivatives share-alike. We therefore:
    • do not redistribute the raw data in the repo (download on demand);
    • credit Olist clearly (see the data dictionary);
    • keep the project non-commercial.
  • ⚠️ Real data is messier (missing values, mixed grains) — handled explicitly in the pipeline, which is itself a demonstrable skill.