VeloDev
/ AI product · WooCommerce

Exact sales reports, explained in plain words

Store owners rarely lack data; they lack time to read it. AI Store Analyst reads orders and stock from WooCommerce, computes the metrics in code, and lets the owner ask questions in plain English or Romanian. The model never does the maths and never sees customer data: it picks one report and explains it, and every number in its answer is checked against that report.

Platform
WooCommerce
Status
Working MVP · open for pilots
Access to the store
Read-only
AI cost per question
~$0.001
Technology Stack
  • Node.js
  • Fastify
  • TypeScript
  • PostgreSQL
  • React 19
  • Vite
  • Claude API
  • WooCommerce REST API
  • zod
/ demo

The dashboard on a demo store with 12 weeks of orders, then a real question to the model. Recorded in the browser and sped up; the answer is the model's own, unedited.

store-analyst / dashboard
Demo
Demo store with generated orders (no real customer data) · store language set to English · Claude Haiku · 1.3× speed
Verified on the demo store
100%
Match with SQL check
12/12
Test questions, EN and RO
0
Customer fields copied
15
Minutes between syncs
Monthly totals also identical to WooCommerce Analytics; the remaining differences (payment date vs order date) are documented.
/ what it does

Numbers you can check

Net and gross sales, discounts, refunds, shipping, tax, orders, average order value and items sold, computed in whole cents in the store's time zone. Each metric shows its formula.

Answers in English or Romanian

Ask about last week, last month, best sellers, falling products or low stock. Questions about profit, customer contacts or forecasts are refused, because the data cannot support an honest answer.

Stock that runs out

Stock cover in weeks, counted per variation, so a T-shirt size that sells fast is flagged before it is gone.

Private by design

The sync never copies names, emails, phones or addresses. Returning customers are counted through a one-way code. Questions and answers are not stored.

/ Engineering snapshot

Challenge

Let a store owner ask questions about sales in plain language without the classic failure of AI reporting: confident, made-up numbers. And keep customer data out of the AI path entirely.

What I owned
  • Product scope from the spec: which questions it answers and which it refuses.
  • Read-only WooCommerce sync, the metrics engine and the report tools.
  • The AI layer: tool selection, number check, cost caps and the evaluation set.
  • The dashboard and the comparison with WooCommerce Analytics.
Result
  • Totals identical to an independent SQL calculation and to WooCommerce Analytics for a full month.
  • 12/12 test questions answered correctly with a real model, in English and in Romanian, including refusals for profit, customer emails and forecasts.
  • A working local MVP, ready for a pilot store.
Key engineering decisions
  1. The model chooses a report; code computes it

    The model gets a closed list of report tools, picks one with parameters (or declines), and explains the structured result. It has no SQL access and no way to compute a number itself.

  2. Every number in the answer is checked

    Numbers are extracted from the answer (English and Romanian number formats and dates included) and matched against the report. If one does not match, the user gets a template answer built from the data instead.

  3. Refunds split the way WooCommerce does

    Full refunds are divided into product, shipping and tax parts. That was the difference that made the totals match WooCommerce Analytics exactly.

  4. Hard cost caps

    Daily and monthly AI budgets are enforced in code; a question costs about a tenth of a cent with Claude Haiku.

/ screens

Dashboard screens (7)

Click a tab to switch, or the image to enlarge.

store-analyst / dashboard
01 / 07

Key figures with formulas

Net sales, orders, average order value, refunds and items sold for any period, compared with the previous one. Every figure opens its formula and notes, e.g. net sales are not profit.

/ how it connects

Sync

  • WooCommerce REST API with a read-only key.
  • Orders, refunds, products and variations every 15 minutes.
  • No personal data stored.

Metrics

  • PostgreSQL, amounts in whole cents.
  • Store time zone, daylight-saving days handled.
  • Written definitions shown in the dashboard.

AI

  • Claude picks one report from a closed list.
  • Number check before the answer is shown.
  • Daily and monthly cost caps.
/ pilot

Try AI Store Analyst on your store

The product works end to end on a demo WooCommerce store; the next step is a pilot on a real one. A pilot is a few weeks on your store with your data, read-only or behind your approval, with the results reviewed together before anything is extended. Agencies can offer it white-label to their own WooCommerce clients.

marian.putirac@gmail.com