Sales Pro
Every purchase, backed by better decisions.
A simple purchasing dashboard for small retailers that helps them know what to buy, when to reorder, and how to check supplier bills. AI gives useful suggestions and explains why, but every purchase is approved by a human.
Role
UX/UI Designer
Team
Solo project
Platform
Responsive web
Tools
Figma, FigJam, Claude Code
View the live demo

The challenge
Stock and supplier payments don’t line up.
Small businesses have extra costs when stock and supplier payments get out of sync. Buyers guess how much to reorder. Accountants retype supplier bills, then check them line by line against orders and deliveries.
I designed for two accountants from different industries, one in retail and one in production, who share the same repetitive routine.
01
Typing supplier invoices into the software by hand
02
Checking each bill against its order and delivery, line by line
03
Guessing how much stock to reorder
04
Duplicate bills arriving through different channels
Problem 01
Buyers struggle to decide what to reorder
A trend line and two stock figures make the answer hard to find. The quantity field starts at a default, so it’s easy to order the wrong amount.
Solution
Explainable restock suggestions
The order form opens pre-filled with a suggested quantity, accepted in one tap, plus a link that shows how the number was worked out, so the buyer can see the reasoning before ordering.

Problem 02
Invoice verification is repetitive and time-consuming
Every month the accountant types supplier bills line by line, then looks for the matching order. It’s slow, and mistakes still happen.
Solution
Three-way invoice matching
The AI reads the bill and matches it to the order and to what was received. Ordered, received and billed sit side by side, with differences marked in red. The work shifts from checking everything to fixing only the bills that don’t match.

Problem 03
Important purchasing issues are spread across screens
Stockouts, overdue payments and late deliveries live in different tables. The user has to scan each one and decide what matters first.
Solution
A Needs Attention dashboard
Issues are ranked into three tiers: Act now, This week and Worth knowing. Each card opens the screen it came from. The home screen answers one question: what do I act on right now?

Problem 04
AI assistants can make incorrect assumptions
A vague request like “create a PO for 30 iphone” could mean three different products. An assistant that guesses picks one without checking with you.
Solution
An assistant that asks before acting
When a request is unclear it asks a short question with answers you can pick. Before anything is created it shows a confirmation card. A switch lets the user decide how much it asks.


The final experience
From the home screen to a sent order.
Selected screens from the 18-screen interactive prototype.

Purchase orders: payment status and delivery progress for every order in one table.

Draft invoices: grouped by how much attention each one needs.

Stock audit: a plain-language question returns a report with a PDF download.
Impact
60%
AI assistance for faster data entry, reducing entry time by 60%.
Prototype and learnings
What I built, and what it can’t do yet.
An interactive prototype
18 documented screens, built with Claude Code and open as a live demo.
A token-based design system
Type, spacing, radius, layers and colour roles are defined once, with light and dark themes.
Tested with two accountants
I put the prototype in front of two accountants, one in retail and one in production.
A units-of-measurement gap
Retail counts in pieces, packs and boxes, which the prototype converts. Production counts in litres, tanks and tubes, which it doesn’t support yet. Next time I’d ask about units in the first interview.
What the AI can’t do yet
The AI uses fixed rules, not a live language model, and the product says so. Data also resets when the page is refreshed.
Project reflection
What I learned building with AI
AI let me build a working prototype without being a developer. To build responsibly, I still need to know enough code, backend and security to question what it gives me. When a padding fix changed half a screen, I had to read the code to see why.
Sales Pro doesn’t save data when you refresh, so a real version needs a database, accounts and real invoice reading. That is the next step I would take.

