Chelo – E-commerce brand site

The messages the brand was already getting said pretty clearly why people came, and it wasn't for a shopping cart. Two needs kept coming back, and only one of them is about commerce.

• Before buying, knowing who you're dealing with. On a marketplace the seller is one line of text. An unknown brand has to prove there is a real company behind it, or the doubt benefits the incumbent.
• After buying, finding an order. The most frequent request by far, and precisely the one the site cannot handle: the order doesn't live there.

Three worlds (Food, Home, Fun) act as the navigational spine. The third hasn't launched, and rather than hide it we say it is being drawn. The visitor reads a brand trajectory, not a catalogue of four references.

Five languages, one URL per language, a selector that stays in the header. They follow the countries where the brand actually sells, not a traffic ambition. One language more than markets would be a promise after-sales cannot keep.

The contact page receives three requests with nothing in common: a partnership, a question about the company, an order problem. The third is the most frequent and it is the only one the site can do nothing about. So it is handled before the form, not after.

• A redirect box at the top of the column sends the buyer back to their marketplace account, where the order actually exists. They are served faster than by writing, and the inbox doesn't fill up with requests nobody can answer.
• The company identity block (registration number, registered office, VAT number) sits opposite the form rather than being relegated to the legal notice. It is what separates “dubious shop” from “identifiable company”.
• The form sends nothing to a server: it opens the visitor's mail client. No database, no collection, no processing to declare.

The production constraint was set in advance: static site, existing hosting, unchanged address. No dynamic component was negotiable, and honestly it framed the interface decisions far better than a specification document would have. The design system had to hold five languages across three worlds, so every component was tested against German first: that's the language that breaks layouts before any other.

Most of this project was done in Claude Cowork, from the analysis right through to going live. Reading through the messages the brand was getting, the decision to drop the checkout, the visual identity, the five-language design system, the interface, the static build and the publication: all of it went through there.

The one stage that sat outside was the wireframes, done in Figma and deliberately so. That is the artefact two people annotate together, and the client needed to see the site map and the templates before we talked about colour. Once the architecture was agreed on screen, the rest came back into Claude Cowork, through to the launch on the original domain.