When there are hundreds of items, going through them by hand takes a long time. The «AI» button on the «Matching» tab opens a block that prepares the data for any chatbot and takes its answer back.
Step 1. Export
Choose the group of your items and the platform category (both are taken together with their subgroups) and press «Export for AI». Two files will be saved:
- JSON — your items and the catalogue items from the program's local copy.
- The prompt text in the language of the session (Ukrainian, English or Russian) — a ready-made instruction for the AI.
Step 2. Hand this to the AI
The prompt and the JSON are sent to any chatbot — ChatGPT, Claude, Gemini. Save the answer to a .json file. A ```json wrapper around the answer is allowed, the data processor recognises it.
In its answer the AI states the confidence and the reason for every pair, and also lists the ones for which no pair was found.
Step 3. Importing the answer
«Import AI answer» checks the file: whether such an item exists, whether the platform item exists, whether there are duplicates. Only confident pairs are ticked — the doubtful ones stay unticked, and you decide on them yourself.
From then on everything is as with ordinary matching: check what is ticked and press «3. Save matches to the program». Until that moment nothing is saved.
A mandatory condition: the producer
For the AI the producer is the key attribute. If it is written in your item list as an abbreviation («KU» instead of «Krono-Ukraine»), the AI will not decipher it and will not find the pair. Enter the abbreviations as aliases on the «Producers» tab — this is done once.
What the AI does not decide for you
- The sale method is always set by the data processor: sheet material — by sheet, everything else — piece / m² / r.m.
- Saving to the program is done by you, with a confirmation.
- The AI's answer is a suggestion, not a fact. What is ticked still has to be checked.