Automatic category matching using artificial intelligence

Brief description

AI-powered automatic category matching feature — is an innovative tool in the Elbuz system that uses artificial intelligence to automate one of the most labor-intensive tasks in e-commerce: matching product categories from suppliers' price lists with the categories of your online store.

What is this function for?

When working with suppliers, you regularly encounter a problem: the categories on their price lists don't match those in your store. For example:

• The supplier uses the "Smartphones and Gadgets" category, while your store uses "Mobile Phones"
• The supplier's category is called "Kitchen Appliances," while yours is called "Kitchen Appliances."
• The same product may have different names: "Laptop", "Notebook", "Portable computer"

Earlier You had to manually create hundreds of matching rules, spending hours on routine work.

Now Artificial intelligence does it for you automatically!

What problems does it solve?

• Time saving: What used to take hours of manual work is now done in minutes.
• Accuracy: AI analyzes not only the names, but also the meaning of categories, taking into account hierarchy and context
• Scalability: Processing hundreds of categories and thousands of products in a single run
• Flexibility: Two operating modes - for categories and for individual products
• Transparency: You can see why the AI chose a particular category and adjust the results.

What will you get in the end?

• ✅ Ready-made import rules, which are automatically applied to all products from the price list
• ✅ Precisely distributed goods according to the correct categories of your store
• ✅ Transparent process with the ability to check and adjust the results
• ✅ Saving 90-95% of time for processing price lists
• ✅ Automation of routine work which will allow you to focus on business development

How the system works

The system offers two operating modes that can be used separately or combined:

Mode 1: Match categories by name

When to use: You have a price list where products are already divided into supplier categories.

How it works:

1. The system loads all categories from the supplier's price list
2. Loads all categories of your store
3. Sends both lists to the AI with instructions to find matches
4. AI analyzes names, takes into account hierarchy (path to category) and semantic meaning
5. Returns a list of matches with a confidence score (0-100%)

Advantages:

• ⚡ Fast: 100+ categories processed in 30-60 seconds
• 🎯 Effective: One matching rule applies to all products in a category
• 💡 Smart: AI understands synonyms and similar words

Example:

Supplier Category: Laptops and computers
AI finds in your catalog: "Laptops" (95% confidence)
Cause: Semantic similarity, laptop = portable computer

Mode 2: Matching categories based on products

When to use: The price list contains products without categories, or the categories are indicated incorrectly.

How it works:

1. The system finds all products without categories
2. Breaks them into bags of 50 pieces
3. For each package, sends a list of products with names and article numbers to the AI
4. AI analyzes each product and determines the most appropriate category
5. The results are saved and applied to the products in the price list.

Advantages:

• 🔍 Precise: Analysis of a specific product, taking into account its characteristics
• 🛠️ Flexible: Works even when there are no categories in the price list
• 📊 Details: Each product receives an individual rating

Example:

Product: "Apple iPhone 15 Pro Max 256GB Space Black"
AI determines the category: "Smartphones > Apple iPhone" (92% confidence)
Cause: Apple brand, iPhone model, memory specifications

Step-by-step instructions for use

Step 1: Open the function

1. Open the section Price lists in the Elbuz admin panel
2. Find the required supplier price list
3. Open the tab Categories

Step 2: Selecting the Analysis Mode

You will see two options on the screen:

Option A: Mode 1 - Matching categories by name

• Use if the price list specifies supplier categories and they have been loaded into the Elbuz program in the "Category from the price list" column (for each product)
• Recommended for initial quick processing of price lists

Option B: Mode 2 - Product-Based Matching

• Use for uncategorized products
• Suitable for "finishing" after Mode 1

Step 3: Run the analysis

1. Select the desired mode (default is Mode 1)
2. Click the button "Run Analysis"
3. The system will begin processing - monitor the progress in the operation log

What happens at this moment:

• Store categories are loading (quantity is displayed)
• Categories/products from the price list are loaded (quantity is displayed)
• A request to AI is being generated
• AI processes the data (usually 30-120 seconds)
• The results are saved in a temporary report.

Step 4: View and check the results

Once the analysis is complete, a table with the results will appear.

Table structure:

Results statistics:

At the top of the screen, cards with statistics are displayed:

• Total: Total number of matches found
• Exact: Number of exact matches
• Similar: Number of similar matches
• Not found: The number of categories/products for which the AI did not find suitable options

Step 5: Filter and adjust the results

Using filters:

There are two drop-down lists above the table:

• Filter by match type: Show all / exact / similar / not found
• Filter by confidence: Any / high (>80%) / medium (60-80%) / low (<60%)

Advice: First, look at the low confidence results - this is where manual adjustments are most often needed.

Changing category manually:

1. Click the pencil icon (✏️) in the store category column
2. In the window that opens, select the correct category from the drop-down list.
3. Click "Save"
4. The category will be updated, the checkbox will be automatically checked, and the confidence will become 100%.

Removing unnecessary mapping:

1. Click on the red cross icon (❌) on the right side of the line
2. Confirm deletion
3. The row will disappear from the table.

Step 6: Bulk Actions

To speed up your work, use the bulk action buttons:

• "Select all": Mark all visible results (taking into account the current filter)
• "Deselect": Unmark all results
• Checkbox in the table header: Check/uncheck everything on the current page

Step 7: Saving the results

1. Make sure all categories you want to save are checked.
2. Click the button "Save"
3. Confirm the action in the dialog box

What happens when you save:

• In Mode 1: The system creates category definition rules that are automatically applied to all products with the corresponding categories. These rules will be applied consistently during subsequent product downloads from the price list—repeated AI analysis is not required unless new categories are added to the price list.
• In Mode 2: The system directly assigns selected categories to specific products in the price list. When reloading the price list, the rules of Mode 2 will not be applied to the products (unlike Mode 1). Recommendation: After saving the category comparison results, bulk add products to the base catalog. They will automatically be assigned to the specified categories and, upon subsequent price list uploads, will remain in the correct categories thanks to the established link with the catalog products.
• Stored mappings are removed from the temporary table.
• Unsaved results remain for further processing.

Step 8: Checking the results

After saving, it is recommended to check:

• Go to the "Manual settings" section (for Mode 1) - new rules for determining categories will appear there
• Open the price list of products and check that the categories have been applied correctly (the "Category" column)
• If necessary, re-import the price list to apply the new rules.

Description of main fields and settings

Basic interface elements

1. Selecting an analysis mode

Mode 1 (Categories):

• Processes a list of categories from the price list
• Creates matching rules
• Fast processing (100+ categories in 30-60 seconds)

Mode 2 (Goods):

• Analyzes individual products
• Apply categories directly to products
• Slower, but more accurate (50 products in 60-90 seconds)

2. Results statistics

• Total: Total number of processed positions
• Exact: Coincidence with 90-100% confidence
• Similar: Coincidence with 70-89% confidence
• Not found: AI did not find a suitable category

3. Results table

Color indication of confidence:

• 🟢 Green (80-100%): You can save without checking
• 🟡 Orange (60-79%): It is recommended to check
• 🔴 Red (0-59%): Manual adjustment required

Match types:

• Exact: The names are almost identical
• Similar: Semantically close categories
• Suggested: AI suggests an option, but is not sure
• No match: No matching category found

4. Operation log

A log of all actions is displayed at the bottom of the screen:

• Launching the analysis
• Processing progress
• AI results
• User actions (mark, delete, edit)
• Saving results

Advice: If you encounter any problems, copy the log and send it to technical support.

Automatic mode settings

Click the button "Settings" (⚙️) to access additional options.

Mode 1 (Categories) Parameters

• Enable Mode 1: Enable/disable category processing
• Minimum confidence for auto-apply: Confidence threshold (default 70%). If the AI's confidence is above this value, the result will be automatically marked for saving.

Mode 2 (Products) Parameters

• Enable Mode 2: Activate/deactivate product processing
• Minimum confidence for auto-apply: Confidence threshold (default 80%)
• Package size: The number of products processed per request to AI (default 50, recommended not to exceed 50)

Automatic mode

⚠️ Warning! Use with caution!

What it does:

• The system automatically launches analysis when importing a price list.
• Results are stored with high confidence without operator intervention
• Saves maximum time, but requires trust in AI

When to use:

• After several successful manual launches
• When you are confident in the quality of AI work on your data
• For verified suppliers with well-structured price lists

Settings for automatic mode:

• Enable fully automatic mode: Main switch
• Set confidence thresholds higher than usual (80-85% for categories, 85-90% for products)
• Results with low confidence will still be saved in the table for manual verification.

Recommendations for filling and use

Getting ready for work

1. Prepare the store category structure:
   • Make sure your store categories have clear names.
   • Structure the hierarchy: "Electronics > Phones > Smartphones"
2. Check the price list:
   • Make sure the price list is loaded into the system.
   • For Mode 1, check for values in the "Category from price list" column.
   • For Mode 2, make sure products have names.

Optimal sequence of work

Recommended order for the new price list:

1. Launch Mode 1 to create rules by category
2. Verify and save results with high confidence (>80%)
3. Manually adjust the results with moderate confidence (60-80%)
4. Launch Mode 2 for uncategorized products
5. Save the results
6. Run price list import to apply all rules

Working with results

High confidence (80-100%)

• ✅ You can usually save without verification
• Review a few results selectively for quality control

Average confidence (60-79%)

• ⚠️ It is recommended to check each position
• Read "The Reason" by AI
• If necessary, change the category manually

Low confidence (<60%)

• ❌ Requires mandatory verification
• Often AI suggests the wrong category
• Change the category or remove the association

No match

• AI did not find a suitable category
• Possible reasons:
  • Your store doesn't have this category.
  • The category name is too specific
  • Category in another language
• Actions:
  • Create a missing category in the store
  • Change the category manually to something similar in meaning
  • Remove the mapping and create the rule manually later

Practical advice

• Use filters: Process the results in stages - first with high confidence, then with low
• Don't rush the first launch: Check all the results to understand how AI works on your data.
• Adjust and retrain: The more manual adjustments you make, the better the system understands your structure.
• Save in batches: You don't have to save all the results at once - you can save them in parts.
• Use the "Update" button: If you need to interrupt your work, the data will be saved and you can return later.

Tips for optimal configuration

Setting confidence thresholds

Choosing the right threshold depends on your category structure:

Optimizing AI Performance

Improving the quality of results:

• Unify category names: Use a consistent naming style across your store
• Avoid abbreviations: "Household appliances" is better than "BT"
• Add category descriptions: Helps AI better understand the purpose
• Structure the hierarchy: "Clothing > Women's > Dresses" is clearer than just "Dresses"

Performance optimization:

• Packet size (Mode 2):
  • Small (20-30) = longer, but more accurate
  • Large (50-70) = faster, but accuracy may decrease
  • We recommend: 50 products (optimal balance)
• Processing large price lists:
  • First, run Mode 1 (will process all categories quickly)
  • Then Mode 2 only for uncategorized products
  • This will save AI tokens and time.

Automatic mode: when and how to turn it on

Conditions for inclusion:

• ✅ You have performed at least 3-5 manual launches at this price point
• ✅ The accuracy of the results is consistently >85%
• ✅ The price structure does not change dramatically from delivery to delivery
• ✅ You understand and trust AI logic

Recommended settings for automatic mode:

• Mode 1: On, threshold 80%
• Mode 2: Disabled (or 90% threshold for very precise work)
• Check the logs after automatic launch

When NOT to turn on automatic mode:

• ❌ This is your first time working with a price list.
• ❌ The supplier frequently changes the category structure
• ❌ There are many similar categories in your catalog
• ❌ You have critical accuracy requirements (e.g., medicine, technology)

Common problems and their solutions

Problem 1: AI doesn't find matches

Reasons:

• The names of categories in the store and on the price list are too different
• Different languages are used
• The store is missing categories

Solution:

1. Check the store's category structure - perhaps new categories need to be created
2. Rename your store categories using more common terms
3. Add synonyms to the titles (e.g. "Smartphones (Mobile Phones)")
4. Use manual matching for categories not found

Problem 2: AI chooses the wrong categories

Reasons:

• Low confidence threshold
• Similar categories in the store (for example, "Laptops" and "Portable Computers")
• Ambiguous names in the price list

Solution:

1. Increase the confidence threshold in the settings (from 70% to 80-85%)
2. Manually check results with 60-80% confidence
3. Combine similar categories in your store or rename them more clearly
4. Use filters to identify problematic results

Problem 3: The request to the AI is too long

Reasons:

• Large number of categories (>500)
• Large package of goods in Mode 2 (>70)
• AI API server overload

Solution:

1. Reduce the packet size in the settings (from 50 to 30-40)
2. Break the price list into several parts
3. Try running it at a different time (there is less load on the API at night)
4. Please contact technical support if the problem persists.

Issue 4: Error "AI did not return a response"

Reasons:

• Token limit exceeded
• No internet connection
• Technical issues on the AI service side

Solution:

1. Check your account balance and limits
2. Try running it in a few minutes.
3. Check your system logs for additional errors.
4. Contact technical support and provide the transaction log.

Problem 5: Results are not saved

Reasons:

• No checkboxes are checked
• The selected results have category_id = 0
• Database error

Solution:

1. Make sure you have checked the boxes next to the results you want.
2. Please make sure that each marked result has a store category listed.
3. For "No Match" results, please select a category manually first.
4. Check the log for errors
5. Try saving the results in parts (10-20 at a time)

Problem 6: After saving, products remain uncategorized

Reasons:

• In Mode 1, rules are created but not applied automatically.
• Price list re-import has not been started.
• The rules conflict with existing ones

Solution:

1. For Mode 1:
   • Open the "Import Rules" section for your price list.
   • Make sure the rules are created
   • Re-import the price list to apply the rules.
2. For Mode 2:
   • Categories are applied directly to products
   • Check the list of products in the price list
   • If there are no categories, try saving again.
3. Please make sure that category rules are enabled in your price settings.

Problem 7: Low accuracy for a certain type of product

Reasons:

• Specific terminology (e.g. spare parts, chemistry, medicine)
• Multifunctional products
• Not enough data for AI (article only, no description)

Solution:

1. For specific categories, use manual matching
2. Create separate direct rules for such products
3. Improve your price list structure by adding product descriptions
4. Use Mode 1 instead of Mode 2 for batch processing

An example of a full work cycle

Scenario: Processing the price list of an electronics supplier

Initial data:

• The price list contains 150 supplier categories
• There are 300 categories in the store
• Total number of products: 5000

Шаг 1: Первый запуск (Режим 1)

1. Открываю прайс → "AI сопоставление"
2. Выбираю "Режим 1: Категории"
3. Нажимаю "Запустить анализ"
4. Жду 45 секунд

Results:

• Найдено 150 сопоставлений
• Точных: 85 (57%)
• Похожих: 48 (32%)
• Не найдено: 17 (11%)

Шаг 2: Обработка результатов с высокой уверенностью

1. Фильтрую "Высокая уверенность (>80%)"
2. Выборочно просматриваю 10 результатов — всё правильно
3. Нажимаю "Выбрать всё"
4. Сохраняю (создано 85 правил)

Шаг 3: Проверка средней уверенности

1. Фильтрую "Средняя уверенность (60-80%)"
2. Просматриваю каждый результат
3. Исправляю 8 неправильных сопоставлений вручную
4. Отмечаю проверенные результаты
5. Сохраняю (создано ещё 48 правил)

Шаг 4: Ручная обработка "Не найдено"

1. Фильтрую "Не найдено"
2. Для 17 категорий вручную выбираю подходящие категории магазина
3. 5 категорий создаю новые в магазине
4. I'm saving

Итого после Режима 1: Создано 150 правил сопоставления категорий за 15 минут работы

Шаг 5: Запуск импорта прайса

1. Перехожу в раздел импорта прайсов
2. Запускаю импорт для применения новых правил
3. Жду завершения импорта

Result: 4500 товаров получили категории автоматически

Шаг 6: Обработка оставшихся товаров (Режим 2)

1. Открываю "AI сопоставление" снова
2. Выбираю "Режим 2: Товары"
3. Нажимаю "Запустить анализ"
4. Жду 5 минут (обрабатывается 500 товаров без категорий)

Results:

• Обработано 500 товаров
• Высокая уверенность: 380 (76%)
• Средняя уверенность: 95 (19%)
• Низкая уверенность: 25 (5%)

Шаг 7: Сохранение результатов Режима 2

1. Отмечаю результаты с высокой уверенностью (380 товаров)
2. Проверяю и корректирую среднюю уверенность (95 товаров)
3. Низкую уверенность обрабатываю вручную (25 товаров)
4. Сохраняю все отмеченные результаты

Result: Все 500 товаров получили категории

Итоговая статистика:

• Общее время работы: 30 minutes
• Обработано товаров: 5000
• Автоматически (AI): 4800 (96%)
• Вручную скорректировано: 200 (4%)
• Time saving: ~8 часов ручной работы

Шаг 8: Включение автоматического режима

После успешной обработки и проверки качества:

1. Открываю "Настройки"
2. Включаю "Автоматический режим"
3. Устанавливаю порог уверенности 80% для категорий
4. Сохраняю настройки

Result: При следующем импорте прайса система автоматически применит AI сопоставление без участия оператора

Additional information

Требования и ограничения

• Лимиты обработки:
  • Режим 1: До 1000 категорий за один запуск
  • Режим 2: До 1000 товаров (автоматически разбивается на пакеты по 50)
• Processing time:
  • Режим 1: ~30-60 секунд на 100 категорий
  • Режим 2: ~60-90 секунд на 50 товаров
• Price: Зависит от тарифа расхода токенов

Безопасность и хранение данных

• Результаты AI анализа сохраняются во временной таблице
• После сохранения результаты переносятся в основные таблицы (правила импорта или товары)
• Несохранённые результаты хранятся до ручной очистки или 7 дней
• Логи операций не хранятся после закрытия окна

Интеграция с другими функциями системы

• Правила импорта: Результаты Режима 1 создают правила, которые можно редактировать вручную
• Импорт прайсов: Автоматический режим интегрирован с процессом импорта
• Product catalog: Результаты Режима 2 напрямую обновляют категории товаров
• Analytics: Статистика по сопоставлениям доступна в логах операций

Feedback and support

Если у вас возникли вопросы или проблемы:

• Скопируйте лог операции из интерфейса
• Опишите, что вы пытались сделать
• Укажите ID прайса и контрагента
• Отправьте информацию в техподдержку Elbuz

Conclusion

Function "Автоматическое сопоставление категорий с помощью AI" — это мощный инструмент, который способен кардинально упростить вашу работу с прайс-листами поставщиков. Вместо многочасовой рутинной работы вы получаете готовые результаты за минуты.

Key benefits:

• 🚀 Speed: Обработка сотен категорий за минуты вместо часов
• 🎯 Accuracy: AI понимает семантику и контекст, а не только буквальные совпадения
• 🔄 Automation: Возможность полностью автоматического режима для проверенных поставщиков
• ✅ Control: Полная прозрачность результатов с возможностью ручной корректировки
• 💰 ROI: Экономия времени окупает затраты на API в десятки раз

Начните использовать уже сегодня:

1. Выберите прайс-лист для тестирования
2. Запустите Режим 1 для быстрой обработки категорий
3. Проверьте результаты и сохраните
4. Оцените экономию времени и качество результатов
5. Настройте автоматический режим для регулярных поставок

С каждым запуском вы будете лучше понимать, как работает система и сможете настроить её под свои специфические задачи. Доверьте рутину искусственному интеллекту и сосредоточьтесь на развитии вашего бизнеса!