↑↓ selectEnter openEsc close

    1. Elbuz

    How to match price list categories to your catalog with AI

    31 min readUpdated

    Elbuz picks catalog categories for supplier categories or products: identical names are matched without the model, AI handles the rest, and you review the pairs before saving.

    When you need this:

    • The price list has hundreds of supplier categories — Mode 1 handles them in two model requests at most and creates distribution rules.
    • The supplier sends a bare product list — Mode 2 picks a category for each product by name.
    • The same categories arrive regularly — automatic mode handles new categories after import.
    • Dozens of suppliers — one run covers all price lists.

    Where to start: open “AI category matching” and run the analysis in Mode 1.

    What AI category matching does

    The window links supplier categories to your catalog categories for the whole price list in one click instead of manual matching pair by pair. Some pairs it finds by exact name, without the model; the rest takes two model requests at most, however many categories the price list has. The window itself promises “100+ categories in 30 seconds”.

    A run produces a report: until you click “Save”, nothing changes in the price list or the rules.

    Which mode to choose

    The rule is short: if the supplier file has a category column, use Mode 1; if not, use Mode 2.

    At the top of the empty window is the “Select analysis mode” switch; Mode 1 is selected by default. The switch disappears as soon as results appear in the table and comes back when you remove them with “Clear”.

    Top of the AI category matching window with the analysis mode selector
    1Mode 1: match categories by name
    2Mode 2: match categories based on products
    3Hint with the time estimate
    1. Mode 1 works from the list of supplier category names and produces a distribution rule.
    2. Mode 2 analyses individual products that have no category and creates no rules.
    3. The info icon next to a mode shows a hint with how long the run takes (Mode 1: 100+ categories in about 30 seconds).
    • “Mode 1: Match categories by name” takes the unique supplier category names from price list rows that don't have a catalog category yet and asks the model which of your categories matches each one. The model gets only the list of names — no products, prices or item counts.
    • “Mode 2: Matching categories based on products” takes price list rows that have neither a category nor a catalog match and asks the model about each product by its name. The description under the switch promises “50 products per batch, ~2 min per 100 products”.
     Mode 1: by nameMode 2: based on products
    What it takes from the price listunique supplier category namesindividual products without a category
    Does the supplier file need a category columnyes, without it the mode won't see a single rowno, it works even when the file has no categories at all
    What “Save” createsa distribution rule — the same as manual matchinga category right in the price list row, no rule is created
    Does it work on future importsyes: the rule picks up all new products in this supplier categorynot for new products, but it gives its own products their category back on every import
    How it runsimmediately, the answer appears in the same windowin the background, in batches of 50 products
    How many model requeststwo at most for the whole price listone for every 50 products

    Mode 1 is an order of magnitude cheaper, and its result lasts longer: the distribution rule will still work next month. If the file has no category column, it replies “No supplier categories found in the price list”.

    You need Mode 2 in two cases: a price list without categories, or what's left after Mode 1 — products from overly broad supplier categories (“Miscellaneous”, “Sale”) and names the model didn't recognize.

    The order “Mode 1 first, then Mode 2” is built into the program: before queuing the job, Mode 2 extends the pairs Mode 1 already found to all products with the same supplier category name. The more Mode 1 did, the shorter the Mode 2 run.

    How to start a run

    Select a mode and click “Run analysis” — the button label becomes “Run analysis (Mode 1)” or “(Mode 2)”. The program asks: “Run AI category matching (Level 1) for this price?” — in dialogs and in the progress bar the modes are called Level 1 and Level 2.

    Control panel: Run analysis, Settings, Manual mode, operation log icon and the Actions burger
    1Run analysis
    2Settings
    3Manual mode
    4Operation log icon
    1. Run analysis: the label changes to Mode 1 or Mode 2 depending on the selected mode.
    2. Settings: automatic mode, thresholds and batch size.
    3. Manual mode: opens the distribution rules window.
    4. The list icon shows and hides the operation log.

    The panel changes with the state of the table:

    • while the table is empty — “Run analysis”, “Settings”, “Manual mode”, the log icon and the “Actions” menu;
    • after a run, “Save” with an arrow, “Refresh” and “Clear” appear, and “Run analysis” is hidden — while the report has rows, you can't start a new run.

    The list icon shows and hides the “Operation log”. During a run the log expands on its own and reports how many categories were matched by exact name without the model, how many rules were saved, and how many products went to the queue. The “Manual mode” button opens the distribution rules window.

    The “Actions” menu has four items: “Ignore categories”, “List of unassigned categories”, “Price list category matching” and “List of assigned categories”. The two lists export to XLSX — handy for sizing the work before a run and checking the result after.

    The Actions burger menu of the AI category matching window, open with four items
    1Ignore categories
    2List of unassigned categories
    3Price list category matching
    1. Ignore categories: supplier categories that are skipped on import.
    2. List of unassigned categories: an export that shows the amount of work before a run.
    3. Price list category matching: goes to manual review of the category tree.

    A second run replies “No supplier categories found in the price list” or shows an empty table. Not a failure: a category already processed by the model comes back after two weeks at the earliest, and if its row is checked — not until you remove the record. The program doesn't pay twice for the same answer.

    To rerun sooner, remove the previous records with “Clear”. “Clear” doesn't touch distribution rules that were already created.

    The three stages of Mode 1

    1. Exact name match, without the model. The program compares the supplier category name with catalog category names, ignoring case, spaces and punctuation: “Vitamin B12 (Cobalamin)” finds “vitamin b12 cobalamin”. A full path match gives 100% confidence; if only the last level matches and the catalog has one category with that name, 95%. If the same name appears twice under different branches, the category goes to the model.
    2. Candidates and the first request. For the remaining categories the program picks catalog categories with similar words and sends them to the model with the list of supplier categories — one request for all of them.
    3. The remainder against the whole tree. Categories without a confident answer go in a second request, with the whole catalog tree. This helps when the supplier writes “Czajniki” and your catalog is in Ukrainian. If the model's own confidence is below 60%, the category stays without a match.
    A supplier category's path from the price list to a distribution rule
    YesNoYesNoSupplier categoriesthat don't havea catalog category yetStage 1: exact name matchwithout the modelDoes the name matcha catalog category?Pair ready: 100 or 95%,type “Exact”, reason“Exact category name match (noAI)”Stage 2: the program pickscandidatesby words in the nameand sends one request to the modelDid the model choose a categorywith 60% confidence or more?Pair in the report: percentage,type and a short explanationStage 3: the remainder goes in asecond requestwith the whole catalog treePair in the reportor a “No match” rowReport: the operator checks thepairsand unchecks the extra ones“Save”: a distribution rule is created,the category is set on price list rowsright away
    • Action or result
    • Decision: the path depends on the answer

    If all categories are resolved at the first stage, there's no model request at all.

    How to read the report and check pairs

    At the top of the window are five tiles: “Total”, “Exact”, “Similar”, “Suggested” and “Not found”. Clicking a tile filters the table. Below are the “Match type”, “Confidence, %”, “Search: source” and “Search: shop category” filters, the “Reset filters” button and the “Matched by filter: N” counter.

    Match results table with the five statistic tiles, columns and rows
    1Statistic tiles
    2Status column
    3Supplier category column
    4Store category column
    5Score column
    6Actions column
    1. Five tiles: Total, Exact, Similar, Suggested, Not found; a click filters by match type.
    2. Status: a ticked box means the record will be saved; some are ticked automatically.
    3. Supplier category: what came in the supplier price list.
    4. Store category: grey path above the category; a hover shows the explanation.
    5. Score: the model's confidence with a match type pill under it.
    6. Actions: the cross removes the row from the report, the ban sign deletes the category products from the price list.

    The table has five columns: the “Status” checkbox, “Supplier category” (“Product” in Mode 2), “Store category”, “Score” and “Actions”. The leftmost column of checkboxes selects rows for bulk actions — don't confuse it with the “Status” checkbox. By default the most confident pairs are on top, a hundred rows per page.

    The “Store category” column shows the path in gray with the category itself under it. The explanation of why the model chose that category is in the tooltip: hover over the store category name.

    The percentage in the “Score” column is the model's own confidence. Below it is a type badge:

    TypeScoreWhat it means
    Exact90–100%The same category. Pairs from the first stage, found by exact name match, also land here.
    Similar70–89%The same category in meaning, with a different name. Worth a look.
    Suggested50–69%The model isn't sure. Always check.
    No match0%No pair was found in the catalog. In Mode 1 the program adds such a row itself so the supplier category doesn't drop out of the report.

    Some categories will stay without a pair, and that's expected: “Clearance” or “New arrivals” have no counterpart in any catalog, and the model doesn't guess what it didn't understand. Such rows are handled by hand with “Change” or left as they are.

    Review and the “Confirmed” checkbox

    The checkbox in the “Status” column is your decision on the pair. Only checked rows go to “Save”.

    Not only “Exact” rows are checked. Everything that reached the confidence threshold and got a category is checked: by default that's 60%, so “Similar” and “Suggested” are included. “No match” rows are never checked. So clicking “Save” right after a run also saves pairs you should have reviewed. The threshold is set in “Settings”.

    • “Activate all” and “Deactivate” apply to all records in the current filtered set — the ones counted by “Matched by filter”, not just the visible page.
    • “Change” to the right of the store category fixes the pair right in the report: the row gets 100%, the “Exact” type, the “Confirmed” status and the explanation “Category changed by operator”.
    • Select rows with the checkboxes in the left column and the bulk actions bar appears: “Set category” (one for all selected rows, also sets “Confirmed”), “Ignore” and “Delete”.

    What happens after “Save”

    The “Save” button has two parts: the button itself and an arrow with a second item — “Save and add products”. If no rows are checked, the program says “Select at least one match”.

    • Mode 1 creates a distribution rule. For each checked row the program creates the rule “supplier category equals this name” and right away sets the category on price list rows that don't have one yet. If a rule for this catalog category already exists, the condition is added to it. The log shows “Saved rules: N”, and the row disappears from the report.
    • Mode 2 creates no rules. It sets the category on a specific price list row. The record stays in the report and on every following import gives the product the same category back — for rows that distribution rules didn't find a category for.
    • “Save and add products” does the same and adds confirmed products with a category but no catalog pair to the base catalog as new items. The log shows “Products added to the catalog: N”.

    A rule created by matching is the same object as after the “Match” button in manual matching. There's no separate “AI link”: the rule works the same way and on every import picks up new products in this supplier category.

    When the price list or catalog changes

    What happenedWhat happens to what you saved
    A supplier category disappeared from the filethe Mode 1 rule stays and simply finds no products; if a category with the same name comes back, the rule picks it up again
    The supplier renamed a categoryproducts with the new name don't fall under the old rule; the next Mode 1 run treats the new name as a new category
    A product disappeared from the filethe Mode 2 record stays; if the same product comes back, it gets the same category, provided the record is confirmed
    A price list product was matched with a catalog productthe Mode 2 record is deleted
    A catalog category was disabledthere will be no new pairs with it, but existing rules and confirmed records keep working
    A catalog category was deletedthe rules for it and all report records with it are deleted; price list rows lose it

    Unconfirmed report rows for a category that's no longer in the file don't disappear on their own — saving such a row creates a rule with no products. Background cleanup deletes unconfirmed records older than two weeks. What happens to catalog products that dropped out of the file is decided by updating the base catalog.

    How Mode 2 runs: products in batches in the background

    You can't switch to Mode 2 right after Mode 1. While the report has at least one row, the mode switch and “Run analysis” are hidden, and “No match” rows stay after saving. Remove what's left with the delete icons or with “Clear” (it doesn't touch Mode 1 rules) — the switch comes back.

    After “Run analysis (Mode 2)”, the program first extends Mode 1 results to products with the same category names, then splits the rest into batches — 50 products by default — and puts them in the background queue.

    The first fork decides whether any model request is made at all. The lower one decides whether the program applies the result itself or leaves the report to you: when started with the button, it's always the second
    NoYesYesNoYesNoMode 2 starts by button,after import or via the chain:Mode 1 pairs at or above itsthresholdare extended to productswith the same supplier categoryAny products left without a categoryand without a catalog pairthat weren't processed in twoweeks?Nothing to process — no modelrequest.When started by button: “There arenouncategorized products to analyze”Products are split into batchesby “Product batch size”and put in the background queueEach batch is one request:product names and the catalog treeCould the model'sanswer be parsed?Product in the report with apercentage and type.Products the model skipped geta “No match” rowThe whole batch is in the reportas “No match”Last batch processedWas the run started automaticallyafter import or via the chainfrom AI product matching?Pairs at or above the Mode 2thresholdare applied without an operatorThe report waits for the operator:rows at or above the threshold arealready checked
    • Action or result
    • Decision: the path depends on the answer

    The model gets only the product name and your catalog tree — no SKU, brand, price or description. The model will sort out “Samsung Galaxy A54” from general knowledge, but a completely vague name gets “No match”.

    • While the run is in progress, a “Level 2 in background:” bar appears under the buttons with the line “Processed: N / M (X%)” and the note “(you can close the window — the job will continue)”. When you reopen the window, it picks up the active job on its own.
    • “Stop Level 2” stops the run; results already received stay in the report.
    • If there's nothing to process, the program replies “There are no uncategorized products to analyze”, and the log shows “✓ All products already matched (Level 1 / cache)”.

    If AI becomes unavailable in the middle of Mode 2, the batch lands in the report as “No match” with the tooltip “parse error” and won't go into a new run for two weeks. Remove these rows with “Clear” and run Mode 2 again.

    How to undo the result

    • Before “Save”, just uncheck “Status”: nothing changes in the price list or the rules. Remove a row with the delete icon, the whole report with “Clear”.
    • After “Save” in Mode 1, the rule already exists. Remove it where all rules live: in the list of distribution rules or in manual matching. Rules carry no “created by AI” mark.
    • After “Save” in Mode 2, change the product's category and be sure to remove the record from the report.
    After “Save” the modes part ways: a Mode 1 row disappears because the rule takes over, while a Mode 2 record stays and gives the product its category back on every import — which is why it has to be removed separately
    run, score below thethresholdrun, score at or above thethresholdcheckbox, “Change”or “Set category”uncheckeddelete icon, “Clear”,background cleanup aftertwo weeksor the catalog category wasdeleted“Save” in Mode 1 —distribution rule created, therow disappearsdelete icon, “Clear”or the catalog category wasdeleted“Save” in Mode 2 —category set on the productevery price list importgives the product thiscategory backdelete icon, “Clear”,product matched with thecatalogor the catalog category wasdeletedUnconfirmed rowConfirmed row“Status” checkedMode 2 recordafter “Save”
    • Start
    • State
    • End

    After “Save” in Mode 2, changing the product's category isn't enough. The report record stays confirmed and brings the previous category back on the next import — as if your edit “reverted itself”. Remove the record with the delete icon in the “Actions” column or with “Clear”.

    With no rows selected, the “Clear” button removes all report records for this price list — from both modes; with rows selected, it deletes only those and reads “Delete selected (N)”. It doesn't touch created rules.

    Two buttons in the “Actions” column

    The “Delete” icon removes the record from the report — the price list and catalog don't change.

    The second button — “Add supplier category to ignore list and remove all its products from the price” — does what it says: all products in this category disappear from the price list; the dialog warns “This action cannot be undone”. It's for categories you don't need at all; to remove a report row, use the delete icon.

    Ignored categories are also skipped on future imports. How ignoring works, the bulk “Ignore categories” window and how to remove a rule: how to remove unneeded categories from a price list.

    How to set thresholds and automatic mode

    The “Settings” button opens the “Auto mode settings” window with six fields. What you set there applies to the current price list.

    Auto mode settings dialog with the automatic mode selector, two mode checkboxes with thresholds and the batch size
    1Enable fully automatic mode
    2Automatic application warning
    3Mode 1 checkbox and threshold
    4Mode 2 checkbox and threshold
    5Product batch size
    1. Enable fully automatic mode has three values; the default follows the global system settings.
    2. The warning says the program applies results by itself, without an operator.
    3. Mode 1 checkbox and its threshold apply only to the automatic run.
    4. Mode 2 checkbox and its threshold; Mode 2 is off in the automatic run by default.
    5. Product batch size is how many products Mode 2 sends to the model in one request.
    FieldDefaultWhat it does
    “Enable fully automatic mode”“As in global system settings”whether to start matching on its own after a price list import; three values — as in the system, enabled for this price list, disabled for this price list
    “Mode 1: Match categories by name”onwhether Mode 1 takes part in the automatic run
    “Minimum confidence for auto-apply (%)” for Mode 160the Mode 1 threshold: it decides which rows are checked in the report and which results are applied in automatic mode
    “Mode 2: Matching categories based on products”offwhether Mode 2 takes part in the automatic run
    “Minimum confidence for auto-apply (%)” for Mode 260the same for Mode 2
    “Product batch size”50how many products Mode 2 sends to the model in one request; from 10 to 200

    The threshold is how far you trust the model: whatever reaches it is checked in the report, and in automatic mode also applied. Many wrong pairs in the report — raise the threshold, starting with this price list. A higher Mode 1 threshold means fewer pairs get extended to products before Mode 2, and more products go to the model.

    The mode checkboxes don't affect the “Run analysis” button — they only decide which modes the automatic run uses. A price list set to “As in global system settings” takes all six values from the program's global settings; an empty threshold is inherited too.

    Automatic mode and the chain from product matching

    The first field has a warning: “Warning! The system will automatically apply the results without operator intervention when the confidence threshold is reached.” With the mode on, after import the program runs Mode 1, applies results above the threshold — creates distribution rules on its own — then queues Mode 2 if enabled. Everything else waits in the report.

    The left branch is the chain from AI product matching, the right one is this window's own automatic mode. They don't both start after one import: if the chain is on, category matching waits for product matching
    YesFinished without errorsFinished with an errorNoYesNoNoYesPrice list import finishedAI product matchingstarts automatically and“After matching, assign categoriesto new products” is on?AI product matchingruns firstAny new itemswithout a category left?Category matchingdoesn't startChain: both modes,regardless of the mode checkboxesand this window's automatic mode“Enable fullyautomatic mode”on for the price list?Modes checkedin “Settings”Mode 1: pairs at or above thethresholdbecome distribution ruleswithout an operatorMode 2: products without a categorygo to the background queue inbatches;after the last batch, pairs at or abovethe threshold are applied without anoperatorEverything below the thresholdstays in the report for the operator
    • Action or result
    • Decision: the path depends on the answer

    By default the mode is off. Turn it on for suppliers who send price lists regularly with an unchanged set of categories.

    Another way to start matching without the window is the “After matching, assign categories to new products” checkbox in the AI product matching settings. Then, after product matching, the program picks categories for uncategorized new items — in both modes, regardless of this window's automatic mode. “Add new products to catalog” skips them, and the chain closes that gap.

    How to process all price lists in one run

    In the “List of price lists”, switch to “All sources” mode, open the “Price list products” panel and choose “AI category matching” in its menu — the item with the globe icon. You can tell the combined mode by the “— All —” badge in the window header.

    WhatOne price listAll price lists
    “Price list / Supplier” column and filternonepresent, with a list of active price lists
    “Settings” buttonpresenthidden — each price list runs with its own settings, which by default are inherited from the global ones
    Starting Mode 1the answer appears right in the windowa background job, one request per price list, with the “Prices processed: N / M” bar
    Starting Mode 2one joba separate job for each price list that has something to process
    Default sortingmost confident on topby price list, and within it most confident on top

    In combined mode, “Save” saves all checked rows across all price lists at once, not just the ones in the filter. Likewise, “Clear” clears the report for all price lists, and “Stop Level 2” stops all active runs. For a single supplier, open the window from its source.

    Scenarios

    A supplier regularly sends a price list with the same categories

    What to do. Run “Mode 1: Match categories by name” once, check the pairs and click “Save”. Then in this price list's “Settings”, set “Enable fully automatic mode” to “Enabled for this price list”.

    What you get. The saved pairs became distribution rules, and every following import sorts products from these categories on its own. After an import, the program runs new category names through Mode 1 itself: pairs at or above the threshold become rules without you, the rest wait in the report.

    What won't happen. Mode 2 doesn't take part in the automatic run without the “Mode 2: Matching categories based on products” checkbox. If the “After matching, assign categories to new products” chain is on in “AI product matching”, this window's own automatic mode doesn't start after import — the chain picks the categories.

    The price list has no categories, but new items need to go into the catalog

    What to do. Select “Mode 2: Matching categories based on products” and click “Run analysis (Mode 2)”. When the run finishes, review the checkboxes and choose “Save and add products” from the arrow on the “Save” button.

    What you get. Checked products get a category in their price list row, and those without a catalog pair are added to the base catalog as new items. The report records give these products the same category back on every following import.

    What won't happen. No distribution rule is created, so products that weren't in the file before won't get a category next time — they need another run.

    New items should get a category on their own after AI product matching

    What to do. In the AI product matching settings, check “After matching, assign categories to new products”.

    What you get. When product matching finishes, the program starts category matching for rows with no catalog pair and no category — in both modes. Mode 1 pairs at or above the threshold become distribution rules, and after the last Mode 2 batch its pairs at or above the threshold are applied without an operator.

    What won't happen. If no new items without a category are left, category matching doesn't start. Whatever didn't reach the threshold waits in the report.

    You have many suppliers, and all of them need processing

    What to do. In the “List of price lists”, switch to “All sources”, choose “AI category matching” with the globe icon in the “Price list products” panel menu and run Mode 1.

    What you get. Each active price list is processed in the background, with progress in the “Prices processed: N / M” line. The report shows pairs from all price lists together, with a “Price list / Supplier” column.

    What won't happen. The filter doesn't limit saving — “Save” applies all checked rows across all price lists, and “Clear” also clears the report for all of them.

    The model makes mistakes more often than you'd like

    What to do. In the price list's “Settings”, raise “Minimum confidence for auto-apply (%)” — for Mode 1, Mode 2 or both.

    What you get. The program will check only pairs at or above the new threshold in the report, and only they will be applied without you in automatic mode.

    What won't happen. The model's answers don't change — only what the program checks and applies on its own changes.

    From the first run to distribution rules

    FAQ

    I ran the analysis again, and the program says there's nothing to process. Why?

    A category or product that has already gone through the model comes back no earlier than two weeks later, and if it's checked — not until you remove the record. To rerun sooner, click “Clear”; it doesn't touch distribution rules that were created.

    Why are some rows in the report already checked?

    The program checks everything that reached the confidence threshold (60% by default), so “Similar” and “Suggested” are included. Review them before “Save” or raise the threshold in “Settings” (checking pairs).

    How do I remove a rule created by matching?

    In the list of distribution rules or in manual category matching — it's no different from one created by hand. In Mode 2 there's no rule: change the product's category and delete the record from the report (how to undo the result).

    Why isn't the category I need among the options?

    The model chooses only from existing enabled leaf categories: intermediate branches, disabled categories and new names aren't available to it, and it can't create categories. Set an existing category by hand with “Change”.

    Can I close the window while a Mode 2 run is in progress?

    Yes, the work continues on the server. When you reopen the window, it picks up the active job and shows the progress. You can stop the run with “Stop Level 2”; results already received stay in the report.

    I click “Run analysis”, and the program says AI is unavailable or the limit is used up. Why?

    “The daily demo limit is used up…” — a €3 daily cap on AI, parsing and media; continue tomorrow or on a paid plan. “No AI access: AI service API key is not configured.” — no service key is set up in the account. In that case neither mode starts.

    Where to find it and who has access

    This window is not in the main menu and has no desktop shortcut: it opens only from price lists.

    From the “List of price lists”
    source → the arrow on the “Product Matching” button → “AI category matching”
    From the source card
    the “Categories” tab
    From the “Price list products” panel
    menu → “AI category matching”; in “All sources” mode — for all price lists
    Access right
    the price lists window; the “Categories” tab — the price list processing rules section
    Prepare
    AI is available in the account: an AI service key is set up; in a demo — within the daily €3 limit
    Header of the price list source with the Product Matching split button and its open menu of five items
    1Caret of the split button
    2AI category matching
    3AI product matching
    1. The caret hides all matching windows, including category matching.
    2. AI category matching is the second item and opens a full-screen window.
    3. AI product matching is the neighbouring feature; it works with products, not categories, and has its own article.

    From the “List of price lists”, the window opens maximized. The “Product Matching” button in the source header has two parts: the right half with the arrow opens a menu, and the item you need is second in it.

    Related topics

    Was this article helpful?

    Missing a feature? Tell us.

    We will add it for free if other users need it too.

    If only you need it, we will agree on it separately.

    Article contents

    Suggest an improvement

    Try Elbuz free