WooCommerce AI: Ask Your Store Questions

An AI assistant is only as good as the data and definitions behind it. Here is what it can read from a WooCommerce store, which questions work, and where it stops.

"WooCommerce AI" usually means one thing: asking your store a question in plain English and getting a number you can trust. That works when an assistant reads real order data and uses fixed definitions for terms like net sales. This post shows what a WooCommerce connection makes available, the questions that suit it, and the limits you should know before relying on the answers.

What an AI assistant needs from a store

Two things. The first is data it can read: orders, refunds, products and customers, refreshed on a schedule so you are not pasting exports. The second is definitions: what counts as an order, what net sales means, which date an order belongs to. Without definitions, two people asking the same question get two answers, and nobody can say which is right. Most of the disappointment with AI on store data comes from the second problem, not the first.

The data a WooCommerce connection provides

The Quiriz WooCommerce connector (in beta, read-only, refreshed daily) imports six datasets from one store. Each connection covers one store and uses a read-only REST key.

• [WooCommerce] Orders: one row per order in every status, with a flag for the statuses Analytics counts, plus country, a first-order flag and a one-way customer key.
• [WooCommerce] Order line items: product, quantity, price, coupons and quantity refunded.
• [WooCommerce] Refunds: amount, items returned, and shipping and tax refunded, on the refund date.
• [WooCommerce] Sales: the table behind net sales, built to match WooCommerce Analytics > Revenue.
• [WooCommerce] Products: prices, stock status and quantity, categories.
• [WooCommerce] Customers: first and last order, orders and amount spent.

Questions that work well

Name the measure and the period, and the dataset follows from the question:

• "Net sales by month this year, and versus last year."
• "What is our average order value by month?"
• "How many new customers did we get each month, and how many came back?"
• "Which products get refunded most?"
• "Which countries do our sales come from?"
• "Which products are out of stock or running low?"

The first is answered from Sales, the second and third from Orders, the fourth from Sales and line items, and the last from Products. A question about a measure the data does not contain, such as ad spend or site sessions, is outside the store and should say so rather than guess. For traffic, see WooCommerce and Google Analytics.

A few habits make the answers better. Ask one measure at a time before combining them, so you can see which part moved. Name the period in full, such as "January through June this year", instead of "recently". When a result surprises you, ask for the rows behind it: which orders, which refunds, which products. A figure that cannot be traced to rows is a figure to distrust, whoever or whatever produced it. Illustratively, if average order value jumps in one month, the useful follow-up is whether orders fell, basket size rose, or one large order skewed the average.

Why stored definitions matter

The answer to "net sales last month" depends on choices: whether returns count on the order date or the refund date, whether shipping and tax are in, which statuses are counted. The connector makes these choices once. Net sales here is gross sales minus coupons minus returns, with returns on the refund date, matching WooCommerce Analytics. Orders count every status except pending, failed and cancelled. Because Company Context stores the definition, the same question asked in January and June uses the same rule, and the answer shows the work so you can see which rows went in.

What is and is not kept about customers

WooCommerce has many guest shoppers, so recognizing a returning guest matters for new versus returning questions. The connector turns the billing email into a one-way key using a secret that is random for each connection. The same email gives the same key within your store, the key cannot be reversed, and it cannot be matched to another store. The email, names, phones, addresses and refund reasons are not copied. The result is behavior analysis, not mailing-list work.

Honest limits

• Order cap. Orders are read newest first up to a cap that an admin sets. Only whole months are kept, so a large store sees its newest months and the answer says it is showing orders since a given month.
• Guest customers. A guest who uses a different email each time looks like several customers. A customer whose earlier orders fall outside the window may look new.
• Mixed currencies. Amounts stay in each order's currency. If you sell in more than one, do not sum across currencies without checking.
• Hosting and firewalls. Some hosts remove the sign-in header, and some security plugins block the request. The connector reports these cases, but fixing them is a change on your side.
• HTTPS and one store. The store must use HTTPS, and each connection covers one store.

None of this replaces checking. Compare net sales for one month with Analytics before you rely on the rest. More examples are at Quiriz for ecommerce, and the dashboard-side routine is in WooCommerce analytics beyond the dashboard. The Shopify version of this post is Shopify AI.

Using it in Quiriz

Quiriz is an AI data analyst built around governed definitions (Company Context) and showing the work. Connect WooCommerce once, confirm that one month of net sales matches Analytics, then ask your questions in plain English. Connectors are read-only, refresh daily and need a paid plan.

Frequently asked questions