How one row in a price list becomes a finished product page

What is wrong with entering products by hand

Price lists arrive as XML, as Excel, or as pages on the supplier's own website. Every supplier has its own category names, its own way of describing the same processor, and its own idea of what "in stock" means.

Someone opens the list, copies the name, hunts for a photo, writes a description, picks a category, works out the price. Five minutes per product on a good day. For a thousand new items that is around eighty-three hours — two working weeks, every week, for something nobody counts as work.

Worse, this work grows with the catalogue, not with the headcount. Every new supplier multiplies it.

What the platform does instead

The import runs in four steps, each of which used to be somebody's manual job.

Reading. Every supplier has its own connector. There are nine so far — Asbis, Dekada, Vali, Polycomp, Also, VZ, Lancom, TechVision, Unics and Xcom. Some read XML, some connect to the supplier's database, some pull the data straight off their website. From the outside the difference is invisible: everything arrives in one common shape.

Recognition. Before a new product is created, the system checks whether it already exists in the store. It compares references, names and specifications and proposes a match. This prevents duplicates — the most expensive mistake in a catalogue, because it surfaces months later.

Writing. AI composes the name, the description, the specifications and the SEO URL. It does not paraphrase the supplier's list; it writes for the buyer. The category is decided automatically too — since March 2026 AI maps the supplier's categories onto the store's own, instead of that being redone by hand for every new list.

Approval. The finished product enters a review queue. The operator sees what AI prepared, corrects whatever needs correcting, and confirms. Only then does the product go to the store. Nothing reaches a customer without a human saying yes.

What it costs

Every AI call is logged — which prompt, how many tokens, what price. The cost is visible per supplier, per prompt type and per day, with budgets and alerts. You know what you are spending before the bill arrives.

What comes next

Import remains the foundation the other modules stand on: prices are watched on the same products, quotes are built from the same catalogue, and the AI assistant advises customers using the same data.

Frequently asked questions

How many suppliers does the system support? Nine are integrated so far: Asbis, Dekada, Vali, Polycomp, Also, VZ, Lancom, TechVision, Unics and Xcom. A new supplier is added as a separate connector, without touching the rest of the system.

Can it work with my price list format? Yes. Connectors read XML, Excel and CSV, connect directly to a supplier's database, or pull the data off their website when there is no other way. The format is handled inside the connector; from there everything arrives in one common shape.

Does a person have to approve every product? Yes, by default. The finished product waits in a review queue and nothing goes to the store without confirmation. That is a deliberate choice — AI prepares, a human carries the responsibility.

What happens when a supplier changes the description of an existing product? On re-push you choose exactly what gets overwritten — price, description, photos, categories — and what stays untouched. Your manual edits are not lost.

Read more: Product Import + AI in detail