E-commerce and retail datasets for analysis

The short answer

Pick by task: transaction and order-level analysis (Olist, UCI Online Retail II), market-basket and reorder modeling (Instacart), review text and product metadata (Amazon Reviews), demand forecasting (M5/Walmart), and search-trend context (Google Trends via BigQuery). All of the sources below are free to access, though a few require a Kaggle account or agreeing to research-use terms before download.

Sources, one by one

1. Olist Brazilian E-Commerce dataset (Kaggle)

The Brazilian E-Commerce Public Dataset by Olist covers about 100,000 real orders placed on a multi-seller marketplace between 2016 and 2018. It is anonymized commercial data — seller and company names are replaced with fictional labels — spread across multiple linked tables: orders, order items, payments, customers, sellers, product categories and written reviews, plus a separate geolocation table mapping Brazilian zip codes to coordinates. It is the most complete free dataset for studying delivery performance, freight cost and post-purchase satisfaction together. License is CC BY-NC-SA 4.0 (non-commercial); a free Kaggle account is required to download.

2. UCI Online Retail II

The Online Retail II dataset on the UCI Machine Learning Repository is transaction-level data from a UK-based online gift retailer covering December 2009 to December 2011: invoice number, product code and description, quantity, unit price, timestamp, and an anonymized customer ID with country. It is smaller and simpler than Olist, which makes it a common choice for teaching RFM segmentation, cohort analysis and basic time-series forecasting. It is free with no login required; UCI publishes it for research and educational use.

3. Instacart Market Basket Analysis dataset

Instacart released an anonymized dataset of over 3 million grocery orders from more than 200,000 users through a Kaggle competition. Its relational structure — orders, order-product line items, product-to-aisle-to-department hierarchy, and an explicit reorder flag — is purpose-built for association-rule mining and next-basket prediction, and it remains the most-cited public dataset for that task. Access is through Kaggle; the license is for non-commercial research and educational use per the competition rules, and a Kaggle account is required.

4. Amazon Reviews (McAuley Lab, UC San Diego)

The Amazon Reviews dataset, maintained by Julian McAuley's lab at UC San Diego, is the largest public source of e-commerce review text and product metadata. The most recent 2023 release includes user reviews (rating, text, helpfulness votes), item metadata (title, description, price, images) and user-item interaction graphs spanning dozens of product categories; earlier 2014 and 2018 releases (143.7 million and 233 million reviews respectively) are still hosted for comparison. It is distributed for academic research use; check the license notice on the specific release page, since terms differ slightly between the 2014, 2018 and 2023 versions, and some require agreeing to a data use form before download. A mirror is also available on Hugging Face under McAuley-Lab/Amazon-Reviews-2023.

5. M5 Forecasting / Walmart sales dataset (Kaggle)

The M5 Forecasting - Accuracy dataset is hierarchical daily unit-sales data for 3,049 products sold by Walmart across stores in California, Texas and Wisconsin, organized into 42,840 series across 12 aggregation levels, with calendar events and pricing/promotion data included as explanatory variables. It is the standard public benchmark for retail demand forecasting because of its scale and hierarchy, and results are directly comparable to a large published body of competition entries. A Kaggle account is required; use is for research and educational purposes under the competition rules.

6. Google Trends data via BigQuery

Google publishes anonymized, aggregated Google Trends datasets in BigQuery, covering the top 25 overall and top 25 rising search queries at daily granularity for a five-year rolling window (US data down to 210 metro areas). It does not replace transaction data, but it is a free way to add search-demand context — for example, checking whether a spike in a product category's sales lines up with a spike in search interest. Access is free through Google Cloud's public dataset program; you pay only for the BigQuery queries you run against it, and a modest free tier covers casual use.

Sources side by side

How to choose

Start from the question you are answering. If you need to reconstruct a realistic order-to-delivery funnel with reviews attached, Olist is the only dataset here that links all of those together in one place. If you want a clean, teachable transaction log for segmentation or basic forecasting, UCI Online Retail II is smaller and easier to work with. If the task is specifically about what people buy together or when they reorder, go straight to Instacart rather than trying to derive basket structure from a generic transaction log. Review and text-mining work should use the Amazon Reviews releases, checking the license on the exact version you pull. And any serious forecasting benchmark should be checked against M5, since it is the dataset most forecasting methods in recent literature already report results on — which makes your own results easier to sanity-check against published baselines.

Before you build a pipeline on any of these, confirm three things: whether the license permits your intended use (several of the sources above are labeled non-commercial or research-use only), how the data was anonymized and whether that affects the analysis you want to run, and whether the time window still matches your question — some of these datasets are several years old and consumer behavior has moved on.

Frequently asked questions

Which dataset should I use to practice market-basket analysis?

Instacart's Market Basket dataset is the standard choice: over 3 million grocery orders from more than 200,000 users, with a clean relational structure of orders, products, aisles and departments and an explicit reorder flag, which makes it well suited to association-rule mining and next-basket prediction without heavy cleanup.

Is the Olist dataset real transaction data or synthetic?

It is real commercial data from a Brazilian multi-seller marketplace, anonymized before release: about 100,000 orders placed between 2016 and 2018, with company and seller names replaced by fictional names. It is one of the few public e-commerce datasets that includes freight cost, delivery performance and written customer reviews alongside the order and payment data.

Can I use Amazon review data for a commercial product?

Check the specific release before assuming so. The McAuley Lab releases (2014, 2018, 2023) are distributed for academic research use; some versions require agreeing to terms before download and are not blanket-licensed for commercial redistribution. Read the license notice on the release page you intend to use rather than assuming an earlier or later version has the same terms.

Can't find a dataset that matches your task? Leave it to us

If none of the sources above fit your exact combination of category, geography and time window, you can give us your task goal and the conditions it must meet, and we run a data availability assessment (3 free credits on sign-up) first — searching reputable data platforms for real and judging matches and gaps against each required item. Even when no perfectly fitting dataset is found, we present the search directions, approximate sources and item-by-item findings honestly for your reference.

Open the dataset search assistant See research services →

New accounts get 3 free searches; every result comes with verifiable source links.

Further reading

• Research services: availability assessment and source verification
• Curated open datasets · structured reference cards