Case study · Retail · Built for a major roofing manufacturer
Portal and online store with an AI consultant for a major roofing manufacturer
The company website, a store on a live catalog, a staff panel and an assistant that answers in text and voice. One system instead of scattered pages, price lists and calls to managers.
How it works: manufacturer catalog, store, staff panel, AI consultant, team
Catalog
products and sections from the old site
Store
filters, cart, request to a manager
Staff panel
drafts, publishing, roles
AI consultant
text and voice, selection from the catalog
Team
manager notifications, analytics
The task
The company has its own production and a sales network. But its website and catalog lived apart from work with customers. Products had to be checked by hand, questions came in by phone, and site materials were updated through a contractor.
We needed to bring the company website, a store on a real catalog and tools for staff into one place. The catalog had to move from the old site without losing edits. Staff had to publish materials themselves. Customers had to get an answer right away, without waiting for a manager to call.
Before
- Website and catalog separate from work with customers
- A contractor updated the site materials
- Customer questions came in by phone
After
- Website, store and staff panel in one system
- Staff publish materials themselves
- The AI consultant answers right away, in text and voice
What we built
A catalog on real data
Products and sections were moved from the old site into a database of their own. A repeat import does not overwrite staff work: descriptions, additions and publications are kept.
Store: search, filters, cart
Sections by product type, filters by price, color and profile, and product cards with stock and unit price. The cart persists between visits. A request to a manager includes a reference price and consent to data processing.
Staff panel
Editors prepare drafts, administrators publish. Roles and permissions by section, an editor for the catalog and news, and versions of the published site with the option to roll back to an earlier one.
AI consultant in text and voice
The assistant answers from the published catalog, picks items for the task and shows product cards. You can talk to it: speech recognition and spoken answers, an "interrupt" button, echo protection, and a pause on silence. Spending on the model and voice is capped by a budget.
Requests and manager notifications
A request from the store or the assistant goes to managers in Telegram through an isolated channel. The manager's reply reaches the visitor, and each request has a receipt that can be restored.
News and analytics
The company news feed comes from the database, with the original dates and images. The panel shows web analytics reports and branded radio statistics. Provider keys stay on the server.
The result
- The website, store and staff panel work as one system. The catalog and news live in the database, not in page markup.
- Customers get an answer right away, in text or voice, and see matching products right in the conversation.
- Managers learn about a request within seconds and reply from a messenger they already use.
- Model and voice costs are predictable: limits are set in the settings, and staff see an estimate of session costs.
A customer account and online payment are the next stage. For now, an order is placed as a request to a manager. Statuses as of October 2026.
What it looks like
A store section: product series and ready-to-ship items in stock with prices. The data comes from a database of their own, maintained by staff 
The catalog with filters, stock and a cart. A floating button opens the consultant on any page 
Customer services on a section page: calculation, consultant, delivery, installation, and sending a project for estimate
Tell us what you do by hand today
or write to us directly