Case study
Small physical retailers - shops selling electronics, tools, equipment, and similar goods - have stock on the shelf, but the online catalog is often empty or out of date. Typing a description for every item by hand (title, description, specifications, SEO metadata) takes too much time, so online visibility and stock turnover suffer.
In a small shop, new products arrive regularly, but every new item on the webshop requires manually writing a title, description, specifications, SEO metadata, and setting a price. When one person does that on top of everything else, the catalog lags behind what is actually on the shelf - customers can't find products the shop really has, and part of the stock stays "invisible" online.
Tokora built a platform (Mobile Planet) where the shop owner photographs a product on a phone and AI automatically generates the complete listing - title, description, specifications, SEO metadata, and a pricing draft - which is published to the webshop, with no manual typing.
The owner or an employee photographs the product on a phone.
AI recognizes the product and its key characteristics from the photo.
AI generates the title, description, specifications, SEO metadata, and a pricing draft.
The listing is published to the webshop, ready for customers.
The platform is designed as a multi-tenant system - each shop gets an isolated slice of infrastructure (database, hosting, image storage), so the same approach can be moved to another shop without sharing data between them. The approach has been proven in production: Mobile Planet actively runs on this architecture, and a second shop (tenant) on the same platform confirms that the same multi-tenant approach can be extended to a new retail partner without additional architecture work.
Questions and answers
AI proposes the title, description, and specifications based on the photo and available product data - it is a high-quality draft, not guaranteed-perfect text. The owner or an employee can review and correct the proposal before publishing, especially for products with specific or rare specifications.
The workflow does include a review before publishing, but the point is that this review takes a minute instead of half an hour: instead of starting from a blank page, the owner gets a listing that is already about 90% done and just fine-tunes the details as needed. With every correction there is also a field where the user can note where AI went wrong; the system uses that feedback as a rule for future generations, so the same mistake doesn't repeat.
It works best for products whose description naturally depends on visual characteristics and technical specifications - for example tools, electronics, equipment, and similar goods. For products where the description depends on information that can't be seen in a photo (e.g. origin, or contents not printed on the packaging), the AI proposal serves as a base to be completed.
