Elbuz finds the name, price, SKU and stock columns in a supplier's file itself: it checks headers against a dictionary and passes unfamiliar columns to AI, which also reads the values. No need to map a new price list by hand.
When you need this:
- New supplier: columns are mapped on the first file; with a typical price list the source is ready at once.
- Unusual headers: the “Auto-map columns” button fills in the rest with AI.
- Unexpected mapping: there are three mechanisms, and you can see which found what.
Where to start: source card → “Columns” section → “Column wizard” → “Auto-map columns”.
What auto-detection can do
- Parsing during file upload: maps a new price list on its own using the header dictionary.
- The “Auto-map columns” button: matches unmapped columns by field names and hands the rest to AI.
- Auto-setup of a new source: calls AI only when the price, or the name and SKU, weren't found.
- Comparison of the three mechanisms: why the button sometimes finds less than file parsing.
Why auto-detection results differ: three mechanisms
There are three auto-detection mechanisms, and they work differently. That's why this can happen: during file upload the app recognized the “Кол-во” column on its own, but the button in the wizard doesn't see the same column.
| Mechanism | When it runs | What it compares |
|---|---|---|
| Parsing during file upload | automatically, and only while the price list isn't mapped yet, including on the first file from email, a link or | column headers against a 142-key dictionary, plus a check of the first data rows |
| The “Auto-map columns” button | whenever you click it | column headers against system field names, the rest against AI |
| Auto-setup of a new source | automatically, when the source has just been created | the same as the button, plus whatever file parsing has already found |
How columns are mapped when a file is uploaded
Parsing runs when the file has just arrived. The app takes candidate rows and checks each cell against the header dictionary.
The dictionary has 142 keys: 67 for spreadsheet formats and 75 for . For spreadsheets a match is found by substring: the header “Артикул виробника (P/N)” is recognized by the “Артикул” part. For XML the match is exact, word for word; or the vendor and vendorCode tags would be confused.
For spreadsheets the dictionary has these words; the app recognizes them in the file header:
| Field | Recognized headers |
|---|---|
| Name | Наименование, Найменування, Товар, Номенклатура, Name, Назва, Название |
| Price | Цена, Price, Опт, Дилер, Ціна |
| RRP | RecommendedPrice, РРЦ, МРЦ, Роздріб, Розница |
| Manufacturer SKU | ArtNumber, Артикул, P/N, Article, , Код произв, Code, ISBN, |
| Supplier SKU | ProductID, Код |
| Stock status (from price list) | amount, Quantity, Наличие, Наявність, Кол-во, Кількість, Qty, Stock, Склад, Резерв, Колво, Остатки, Залишок |
| Manufacturer | Vendor, Производитель, Виробник, Бренд, Brand, Марка |
| Price category | category, CategoryName, Group, Категория, Категорія, Группа |
| Model | Модель |
| Short description | Описание, Опис, Description |
| Weight | Вес, Вага, Weight |
| Warranty | Гарантия, Гарантія, Гар (мес) |
Four more words, “Валюта”, “Упаковка”, “Размер” and “Заказ”, don't assign fields. They mark the header: the row with the most such hits is treated as the header row.
One field goes to one column. Once a field is taken, later columns with it are skipped; otherwise “ЦенаГРН”, “ЦенаРЦ” and “Цена промо” would all three become the price. The leftmost column wins. If a header matches keys of several fields, the field whose key comes first in the dictionary wins.
For spreadsheets the app then checks the first data rows, in two cases that would otherwise be mapped wrong:
- There's no name column, but there is a “Model” column. The app builds the name from the manufacturer and the model with the address
C,{ },D. If the model in the file already starts with the manufacturer (“Aiwa JH24DT180S”), the model itself becomes the name, with no “Aiwa Aiwa …”. - The “Артикул” (SKU) column contains only digits, and there's a separate “Model” column next to it. It's the supplier's internal code, so it becomes the “Supplier SKU”, and the model the “Manufacturer SKU”. A model with a brand inside doesn't become the manufacturer SKU: catalog matching wouldn't work with it.
A price list arriving by email, via a link or from FTP is mapped on the first file received if there are no column addresses yet; no separate sample file is needed. If mapping fails, the run stops with the message “Column addresses for loading data from the price list are not filled in!”.
For XML there's one more check, by data rather than names. A tag where over half the values look like image links becomes the photo list tag. A tag whose values match category codes from the same file is linked to the category tree.
One setting turns it off: “Detect columns automatically” in the “Price lists” section, “Import and parsing” group, of the Account → System settings window. It's on by default.
1
Automatic column detection switch- Automatic column detection switch — it is on in this account; the system finds the name, price, SKU and stock columns by itself.
The second condition matters more: parsing works only on a new price list. For an Excel workbook and CSV, until the source has saved column names; for XML, also until no field has an address.
The double condition for XML came after an auto-added “Currency” column switched a hryvnia price list to dollars and multiplied all prices by the exchange rate. For this mechanism doesn't exist at all.
Hence a useful trick: if the mapping is hopelessly tangled, clearing the addresses and re-uploading the file is cheaper than fixing it row by row. On a clean price list auto-detection runs from scratch.
What the “Auto-map columns” button does
The magic wand button sits at the top of the wizard window. It doesn't do the same thing as file parsing.
The button doesn't use the 142-header dictionary. It takes the header of each column that isn't mapped yet and compares it with system field names: substring match in either direction, case-insensitive.
A “Product name” column will find the “Name” field, because the field name is contained in the header. But a “Кол-во” column finds nothing: there's no field with that name in the system; it's called “Quantity”. During file upload the same column would be recognized, because “Кол-во” is in the dictionary.
What isn't matched by name goes to AI, but only if columns with headers are left. Meanwhile you see “AI is analyzing the price list…”; it takes up to a minute. Then comes “AI matched fields:” with a number. It's a total: it includes pairs found by field name, not only AI's.
AI receives the list of all target fields, including your custom ones, and the list of columns: ID, header and up to four sample values truncated to 90 characters. So it looks not only at the column name but also at what's in it; that's its main advantage over the dictionary.
The server checks the response: anything not in the lists is discarded, and AI can't take the same column twice.
Neither step breaks anything. The button doesn't touch columns you've already assigned a field to, and won't give a taken field to anyone else. So it's safe to click it halfway through: it fills in what's left and leaves your part as it is.
- Start or end
- Action or result
- Decision: the path depends on the answer
Without AI for the account (say, a demo), the button applies only name matches and shows their count or “Could not match automatically”, then a red message with the reason: in a demo, “AI returned an empty response”; without access to the price list window, “AI suggestion error”. If AI answered but found no pairs: “AI found no matches”.
How a newly created source is set up
You see auto-setup once, when the source has just been created. It also runs on the step from the second to the third step of the connection wizard. The order is: first, whatever file parsing found is taken into account, then headers are compared with field names.
Then one condition decides: if the price wasn't found or a key field wasn't found (name or SKU), AI is called. If both the price and a key field are found, AI isn't involved at all. That's what happens with a typical price list, so there's nothing to wait for.
The gate after which the app considers the source ready is simple: there's a price and there's a name or SKU. If not, the source card shows “Could not detect price or a key field. Please check the column mapping manually.” and the “Configure columns” button.
- Action or result
- Decision: the path depends on the answer
The process itself shows in the card as four steps under the heading “Setting up the source automatically”: “File format”, “Column mapping”, “Loading products”, “Done”. For JSON the second step does nothing: there's no auto-detection for this format at any of the three levels.
Scenarios
A new price list with familiar headers
The supplier sent a spreadsheet with “Name”, “Price” and “SKU” columns. Don't click anything: file parsing finds the price and a key field, and the source becomes ready without AI. Result: in the “Columns” section the addresses are already filled in; all that's left is to check the mapping.
Columns with no header or labeled “Кол-2”
Neither the dictionary nor the field name comparison will recognize such a column. Open the wizard and click “Auto-map columns”. Result: AI gets sample values and assigns a field based on them; it doesn't touch columns you've already mapped.
The mapping is hopelessly tangled
Clear the column addresses and upload the file again. Result: the price list is treated as new again, and file parsing maps it from scratch using the dictionary.
What to do when auto-detection didn't work
| What you see | Why | What to do |
|---|---|---|
| “Could not match automatically”, sometimes followed by a red “AI found no matches” | Auto-detection found no new pairs: the headers don't resemble system field names, and AI produced nothing. A second case with the same text: AI responded, but all its pairs are already taken | Check whether the field you need is assigned to another column (one field goes to one column); then map by hand, starting with the name, price and SKU (how the button works) |
| “AI suggestion error”, “AI returned an empty response”, “Could not parse the AI response” | The AI request failed or the response couldn't be parsed. “AI suggestion error” also appears when you don't have access to the price list window | Try again later. Pairs found by field name are already applied and won't be lost |
| “Price file not found or not uploaded” | Auto-detection was called, but the source has no file | Upload the file in the “Source” section of the card (uploading a price list) |
| “Could not detect price or a key field. Please check the column mapping manually.” | Auto-setup of the new source didn't find the price and/or the name or SKU | Click “Configure columns” and map what's missing (source auto-setup) |
FAQ
I clicked “Auto-map columns” and the result is worse than it was after file upload. Why?
These are different mechanisms. File parsing compares headers against a dictionary that includes “Кол-во” and “P/N”. The button compares a header with a system field name and hands the rest to AI. More: three mechanisms.
I uploaded a new file to an already mapped source. Why didn't auto-detection work?
File parsing works only on a new price list, until the source has saved column names. For a source that's already mapped, click “Auto-map columns” in the wizard. More: parsing during upload.
How long does it take AI to map the columns?
Up to a minute. While AI works, the message “AI is analyzing the price list…” is shown, then “AI matched fields:” with a number. More: the “Auto-map columns” button.
Can I turn off auto-detection?
Parsing during upload is turned off with the “Detect columns automatically” switch in system settings, “Price lists” section. The button in the wizard works only when you click it.
Why does auto-detection do nothing for JSON?
There's no auto-detection for JSON at any of the three levels; mapping is fully manual. More: mapping JSON.
Where to find it and who has access
- Menu
- Directories → Processing price lists → List of price lists
- Button
- source card → “Columns” → “Column wizard” → “Auto-map columns”
- Turn off
- Account → System settings → “Price lists” → “Detect columns automatically”
- Access right
- the “List of price lists” window in group permissions
- Prepare
- an uploaded price list file: without it there is nothing to compare
Related topics
- If you need the big picture of column mapping, see Price list column mapping.
- If auto-detection got it wrong and you need to fix fields by hand, see How to map columns in the wizard.
- If the price list is in XML or JSON, see How to map an XML, YML, JSON and CSV price list.
- If you're deciding whether to hand price lists over to the app, see what it gives your store.
