AI Attributes

Generate attribute, category, and content suggestions from product images using AI; apply the approved ones to PIM.

AI Attributes screen lets you analyze product images with AI to generate suggestions for attributes, categories, and content (product name, description). The catalog team uses this screen to fill missing attributes, link the product to the correct category, and improve the product name or description. Whatever AI suggests, nothing changes automatically in PIM: only the suggestions you mark and approve with Write selected to PIM are written to the product.

You can run the screen instantly for a single product or start a bulk job for many products and monitor progress live. Each AI call per product is chargeable; cost is shown on the screen.

Accessing the screen

  • Follow the path PIM & Catalog › Operations › AI Attributes in the left menu.
  • The AI Attributes button at the top of product details opens the screen for that product’s option code (or SKU if no option code). If neither exists, the button is hidden. If AI was run before for the product, the last result is shown directly. The number on the button indicates how many items were previously transferred from AI suggestions to this product.
  • When you select products on the Products or Sales status screens and start the AI attribute suggestion bulk action, you are redirected here and the job opens with live tracking.

If you don’t see the menu, your account lacks permission for this screen. If you see the screen but the Run, Bulk, and Write selected to PIM buttons are disabled, you lack product write permission; hovering over Run shows "Product write permission is required to run."

Screen sections

  1. Run card: Code field, source, model, auto-approve, and category settings; on the right, Run, Bulk, and Card / List view selection. Pressing Bulk opens the Bulk run section below the card.
  2. Progress panel: Visible during a job; counters, cost, live product flow, and error distribution.
  3. Write result panel: Shows item-by-item results after writing.
  4. Result cards: For each product, images, category suggestion, content suggestion, and attribute table.
  5. Mapping panels: Unmapped attributes and Unmapped categories (visible only if needed).
  6. Write bar: Sticks to the bottom of the page; shows selection count and Write selected to PIM button.

If no results yet, the center shows "No results to display." and a prompt to enter an option code or SKU and run.

Run settings

Field Purpose Tip
Option code / SKU Field to enter the product’s option code or SKU for single product run. Pressing Enter here is the same as clicking Run. If the code matches multiple products, a note "… matches, first used: …" appears below and the first product is used.
Category / attribute source Determines from which catalog AI will get category and attribute lists. Options are in "CODE (… category)" format; the first catalog in the list is selected by default. If no catalog, shows No catalog found; if list cannot be loaded, shows Catalog list could not be loaded.
Model AI model to use. Approximate cost per product is shown for each option. Default is the first and lowest-cost model in the list.
Auto-approve confidence Suggestions above this confidence threshold are auto-approved and pre-checked when results open. Options: 95% (default), 90%, 85%, and Off. Selecting Off means no suggestions are pre-checked.
Consider the current category Whether AI should consider the product’s current PIM category. Unchecked by default.

These settings apply to both single and bulk runs.

Operations

Running for a single product

  1. Enter the product’s code in Option code / SKU.
  2. Check source, model, and auto-approve settings.
  3. Click Run (or press Enter). The button shows Running… and is disabled while running.
  4. The AI is inspecting the product panel opens with a note "This may take 10-30 seconds." The panel shows steps: preparing images, sending to AI, extracting image signature, searching candidate categories, verifying category, reading category attribute list, and extracting attributes. Steps progress with approximate times; the last step remains open until results arrive. Do not press Enter again during this time; each run is a separate, chargeable AI call.
  5. Result card and Progress panel open (for single runs, the panel loads once and does not update live). The code is added to the page URL; refreshing or returning from another screen reopens the last run for the same product.

If the code is empty, a warning "Enter option code or SKU." appears. If no product matches the code, a notification "No product found for this code: …" appears and "No match." is shown below the field. If source cannot be selected, a "No catalog found" warning prevents running. If AI returns no results, "No results to display." appears in the center.

Bulk run

Use when you want suggestions for many products.

  1. Click Bulk on the run card. The Bulk run section opens. Model, source, and auto-approve settings are taken from the fields above.
  2. In Product selection, choose one:
    • Products with images that AI has not processed yet: Processes up to Product limit products with images that have not had AI run before.
    • The code list below: Processes codes entered in the Option codes field.
  3. Enter the maximum number of products to process in Product limit (between 1 and 500, default 50). This field applies only to the first option.
  4. If code list is selected, enter codes one per line or separated by commas/semicolons in Option codes.
  5. Click Start bulk run.
  6. In the Start a bulk AI run? box, review how many products will be sent to which model. The box reminds that each product call is chargeable, the job can be canceled after starting, but costs for completed calls are non-refundable.
  7. Confirm with Start or close with Cancel.

When the job starts, the bulk section closes, previous results clear, and the progress panel opens. The job ID is added to the page URL; refreshing keeps tracking at the same point. If the code list is empty, a "Code list is empty." warning appears; if no products match, "No products matching criteria found." appears and the job does not start.

Monitoring, canceling, and retrying errors

The progress panel updates every 2 seconds during bulk jobs. Live flow replaces result cards while running.

  • The header shows Progress, job number, and job status.
  • Below the progress bar are counters for Total, Completed, Failed, Pending, Processed, and cost in dollars.
  • Live flow titled "AI is processing … product" shows each product’s code, name, and status: queued, processing, Completed, Bug, Skipped. If an error occurs, the error message replaces the status.
  • If errors exist, the Errors section expands with a summary "… / … products failed." Errors are grouped by message; each group shows a "… products" badge and product codes. Clicking a code opens the product in a new tab. If more products exist than shown, "and … more" is displayed.

Cancel stops the job if it is not finished and cancellation was not previously requested. Costs for completed calls are not refunded.

Retry failed appears only if there are failed products and retries them; tracking restarts from the beginning. This button does not ask for confirmation and each retried product is a new chargeable call; do not press repeatedly without reading error messages.

If you lack product write permission, Cancel and Retry failed are disabled.

When the job finishes, result cards for completed products load.

Reviewing suggestions

Each product has a result card.

Card header: Product name, icon to open product in a new tab, and below, option code, SKU, and barcode. On the right, a "category %…" badge shows AI confidence in the category suggestion: green for 85%+, yellow for 60%+, red below 60%; if no suggestion, "no category" is shown. The checkbox (This whole product) toggles selection of all transferable items for the product.

Images: In Card view, all product images appear on the left; in List view, the first four images appear small. Clicking an image opens the Product image window: zoom with mouse wheel or buttons, double-click, drag when zoomed, navigate images with ← → keys, close with Esc. Unloadable images are hidden.

Category section:

Row Meaning
AI suggestion AI’s suggested category path in the selected catalog, source code, and category ID.
Current category Product’s linked category in PIM; if none, "no linked category."
PIM counterpart PIM counterpart of the suggested category and a checkbox. Checking it links the product to this category on write. If already linked, the box is disabled.

If multiple PIM categories match the suggested category, a selection list and note "This source category matches … PIM categories; select one." appear. If no PIM counterpart is mapped yet, a yellow warning appears; you must map it first to link the product. If AI cannot suggest a category, "AI could not suggest a category." is shown. Edit the mapping and Open the category mapping screen links open the Category mapping screen in a new tab; these links appear only if you have permission.

Content suggestion section appears if AI suggested product name or description. Each line shows suggested text in green and current value crossed out below (or "no current value" if none). If suggestion matches current value, the checkbox is disabled.

Attribute table:

Column Meaning
(checkbox) Marks the suggestion for writing. The header checkbox selects all applicable suggestions. Hovering over a disabled box shows why it cannot be selected.
Property Attribute name. Required attributes show a Required badge; those without a PIM field show "No PIM counterpart."
Current value Currently registered value in the product.
AI suggestion Suggested value. Green: new and transferable; yellow: no PIM counterpart; faded: no value or same as current.
Status Whether the suggestion matches allowed value lists (see table below).
Confidence AI confidence in the suggestion (percentage).
Evidence AI’s explanation supporting the suggestion.

Extra findings not in the category’s attribute list are shown with a different background. Below the table, "Current attributes without AI suggestions" lists attributes present in the product but untouched by AI; these are unaffected by writing.

In Card view, results show one by one; if multiple products, navigate with Previous product / Next product arrows and "… / …" counter. In List view, all products are listed vertically. The chosen view is saved in the URL.

Writing selected to PIM

  1. Mark the attributes, content suggestions, and category links you want to write. To select all applicable items across all products, use the Select applicable button or the Select all checkbox in the write bar. Pressing Select all again clears all selections.
  2. Check the "… items selected" count in the bar.
  3. Click Write selected to PIM.
  4. The Apply the selected suggestions to PIM? box shows how many suggestions will be written and that existing product attributes, categories, and content may change. Confirm with Apply or close with Cancel.
  5. The Sending selections to PIM box opens, showing steps: verifying items, mapping to PIM fields, writing category links, and sending to product queue. This box cannot be closed or interrupted.
  6. At the end, a notification Synced or if errors, Partially written appears with a summary "… sent · … skipped · … errors."

After writing, the Write result panel opens. Each row shows the item’s result code (red Failed if error, gray Skipped if skipped, green badge if successful), product code, attribute or content name, written value, and any error message. If errors exist, you can retry only those with the Retry failed (…) button. Suggestions reload with updated statuses after writing.

If no items are selected, Write selected to PIM is disabled. While writing, the button shows Writing…. Retry failed (…) rewrites only failed items without confirmation; the result panel updates with new results.

Values written to PIM cannot be undone from this screen. To revert, edit the product in the Products — detail screen.

Mapping unmapped attributes

If AI extracts attributes without corresponding PIM fields, these suggestions cannot be written. The Unmapped attributes panel aggregates such attributes from all products in one row: if the same attribute appears in 40 products, it shows once and one mapping solves all. Rows are sorted by the number of products descending.

Each row shows the attribute name, "found in … products" info, and up to three example values.

  1. Select the PIM field to write the attribute to from the Pick a PIM field list. If a similar field is found, a button "Did you mean: … (%…)" appears below; clicking it selects that field.
  2. Click Map.

If no suitable field exists, click Create new field. The opened section pre-fills the field name with the attribute name; a new text-type, product-level, active attribute field is created in PIM and linked to the attribute. Edit the name if needed and click Create and map or close with Cancel. Name cannot be empty.

The saved row remains visible with confirmation like "attribute → field (rule saved)" or "(new field created)". Suggestions reload and items now transferable become selectable in the table.

Mapping is permanent and saved without confirmation: from now on, this attribute automatically maps to that field in all future runs. The Saved mappings section below lists these rules as "attribute → field"; it is visible even if no unmapped attributes exist. To remove a rule, click the cross next to it; confirm with Remove mapping in the Remove this mapping? box. The box reminds that removing the rule stops automatic writing of this attribute in future runs.

If the PIM field list cannot be loaded, the panel shows "PIM field list could not be loaded." Retry with Refresh.

Refresh reloads the PIM field list and saved mappings; it does not refresh suggestion tables. If all attributes have counterparts, the panel shows "All attributes in this run have a PIM counterpart." Mapping and field creation buttons are disabled if you lack attribute write permission; the remove cross is hidden.

Mapping unmapped categories

If AI’s suggested category has no counterpart in the PIM tree, the product cannot be linked. The Unmapped categories panel lists these source categories one by one. Each row shows the category path, "found in … products" info, source code, and category ID.

  1. Select the target PIM category from the Pick a PIM category list. If a similar PIM category is found, a "Did you mean: …" button appears.
  2. Click Map.

Mapping is permanent, saved without confirmation, and used for all products from this source category. If the source catalog is not defined in the setup, an error "Source catalog is not defined in this setup, so mapping cannot be saved." appears. If the PIM category list cannot be loaded, the panel shows "PIM category list could not be loaded." The saved row appears confirmed as "source → PIM category" and suggestions reload. The Open the category mapping screen link opens the Category mapping screen in a new tab for detailed mapping. If you mapped categories in another tab, refresh this panel with Refresh to update category lists and suggestion tables. Map is disabled if you lack category write permission.

Statuses

Status badges in the attribute table:

Status (on screen) Meaning Next step
in list Suggestion matches a value in the allowed list. If not exact, match type is shown. You can review and write the value.
outside list · marketplace accepts it Suggestion is not in the list but the channel accepts free values. Review carefully and write if correct.
low confidence · needs review AI confidence is low. Check the image and mark manually if correct.
outside allowed list Suggestion is not in the allowed list. Cannot be written as is; enter manually in product details.
not found in the image AI did not detect this attribute in the image. No value to write.
extra finding Additional info not in the category’s attribute list. If no PIM counterpart, map to a field in the mapping panel.

If a suggestion’s checkbox is disabled, hovering shows the reason:

Reason (on screen) Meaning
This value has already been written to PIM. Suggestion was previously written.
No value to write. AI did not generate a value for this attribute.
No corresponding field in PIM. Map to a field below. Attribute is not yet mapped to a PIM field.
Suggestion matches the product’s current value. Writing would not change anything. Case and whitespace differences do not count as changes.
Suggestion is not in the allowed value list. Cannot be written as is. Disabled to prevent incorrect data entry.

Common scenarios

  • I came from product details, it says "No AI run has been done for this product." No saved results exist. Check settings and click Run.
  • I cannot check the suggestion box. Hover over the box to read why. The most common reason is unmapped attributes; map them in the Unmapped attributes panel to enable the box.
  • No suggestions are pre-checked. Auto-approve confidence may be Off or suggestions fall below the threshold. Mark suggestions manually.
  • There is a category suggestion but I cannot link the product. The suggested category’s PIM counterpart is not mapped. Map it in the Unmapped categories panel. If the product is already linked to that category, the box is disabled.
  • Bulk job gave errors on some products. Check error messages and product codes in the Errors section, then click Retry failed.
  • I refreshed the page, did results disappear? No. The code or job ID is saved in the URL; refreshing reopens the last run or job.

Related screens