Sell your e-commerce data to AI companies

Every order an online shop changes, cancels or refunds under its own policy is the kind of task AI companies now build and test shopping agents on. Years of orders, returns, supplier threads and customer conversations make useful e-commerce data for AI training. Card data and shopper details come out first, and data from marketplaces such as Amazon is bound by the platform's terms.

  • Industry
  • By
    Keilian Knudsen
  • 4
     min read
  • Updated
    20 Oct 2026
ASCII illustration of stacked shipping parcels

Key takeaways

  • In τ-bench's retail test, GPT-4o's chance of getting the same order task right eight times in a row fell to about 25%.
  • When we review a store's records, connected ones count most: an order linked to the shopper's message, the refund decision and the stock change it caused.
  • Card data never goes in, and Amazon's and Shopify's terms limit what their sellers and apps can share.

Which store records interest AI companies?

e-commerce and retail records
RecordWhat it showsWhat comes out first
Order and fulfilment historyWhat was bought, and how orders were split, delayed, changed or cancelledShopper names, addresses, emails, phone numbers and traceable order numbers
Returns, refunds and exchangesHow a returns policy gets applied, including exceptions and goodwill refundsShopper details, photos customers sent, and payment references
Customer service conversationsSizing questions, "where is my order" chases and complaints, and how each one endedAnything shoppers wrote about themselves, and pasted card numbers
Product catalogue and listing editsHow products get described, priced and categorised, and rewritten when they don't sellBrand photos and copy you don't hold rights to, and marketplace content
Supplier and purchasing threadsHow a buyer negotiates minimum orders, chases a late container or switches factorySupplier contacts, and prices or terms agreed in confidence
Stock, promotion and season planningWhich discounts ran, how seasonal stock was bought and what sold throughCustomer-level marketing lists and ad tracking data

Online retail is a testing ground for AI agents

Online retail is where AI agents are already being tested. τ-bench, a research benchmark from June 2024, gives an agent a simulated shop with 1,000 orders and a written policy, then asks it to change, cancel and return orders. GPT-4o solved about 61% of tasks, but its chance of getting the same task right eight times in a row fell to about 25%.

AI companies are also building shopping into their own products. OpenAI launched Instant Checkout in ChatGPT in September 2025, starting with US Etsy sellers, and its GDPval benchmark counts retail trade among its nine sectors, testing first-line supervisors of retail sales workers.

Where does a store's history actually sit?

A store's records usually sit across a commerce platform, a help desk, email marketing and an inventory app or ERP.

  • Shopify
  • BigCommerce
  • Adobe Commerce
  • WooCommerce
  • Amazon Seller Central
  • Gorgias
  • Zendesk
  • Klaviyo
  • ShipStation
  • NetSuite

Card data, shopper details and platform terms

Retail data answers to three sets of rules: the card industry's PCI DSS, privacy law for shopper data (including the GDPR for EU shoppers), and the terms of the platforms you sell through.

  • Card data of any kind. PCI DSS bars keeping card verification codes after authorisation, even encrypted.
  • Shopper identities. Names, addresses, emails, phone numbers, logins and browsing histories.
  • Amazon customer data and Amazon materials. Amazon's seller agreement bars using Amazon materials to develop machine learning models, with a carve-out for your own non-personal sales figures.
  • Exports made through Shopify apps. Shopify's API terms bar app developers from training AI models on merchant data, even anonymised.
  • Other people's content and terms. Brand photos and copy used under licence, and supplier pricing agreed in confidence.

Connected records beat big exports

Retail records are worth most when they connect: an order linked to the shopper's message, the refund decision and the stock adjustment it caused shows the whole job. Years of history across peak seasons and policy changes add depth, as do judgement calls such as fraud holds and supplier disputes. Your own help desk and supplier email raise fewer platform questions than marketplace exports. If you go ahead, you decide which sources, date ranges and uses are included before you sign. It's a licence rather than a sale, so you keep the originals, and selling company data to AI companies explains the difference.

Frequently asked questions

Not the customer data itself. Shopper names, contact details and card data stay out. What a store can license is the record of how it runs: orders, returns, support conversations and supplier threads, with personal details removed. Shopify's API terms stop app developers training AI on merchant data, so check how any export is made. Our guide on whether licensing company data is legal covers the wider checks.

Mostly no. Amazon's Business Solutions Agreement treats information identifying Amazon customers as Amazon's confidential information and bars using Amazon materials to develop machine learning models. It carves out your own non-personal sales data: products, prices, volumes and dates. Supplier emails and internal decisions sit outside it. Have a lawyer read your agreement.

PCI DSS governs card data, not order records. It bars keeping sensitive authentication data, such as card verification codes, after authorisation, and requires stored card numbers to be protected. So card data must be absent. Orders, returns and support threads can then be considered, subject to privacy law and your contracts. The free check tells you whether yours could qualify.

Sources
7

Keep reading

ASCII illustration of a support headset

Sell your customer support data to AI companies

Tickets, chats and escalations show how real problems get solved. See why AI companies want support history, what comes out first, and check yours.

Abstract rows of white points of different lengths on black, like lines of records

What company data do AI companies buy?

Slack and email threads, CRM history, support tickets, project tools and documents. See which business records AI companies want, and why depth counts.