Automatically populate and enrich product cards using AI

What is the automatic product filling and enrichment system?

Imagine: You have thousands of products in your catalog, but many of them have blank product cards—no specifications, descriptions, or images. Filling them out manually takes weeks of monotonous work. But now everything has changed!

Elbuz Automatic Product Card Filling System — is a smart assistant that does all the work for you. You just press one button, and the system:

• Searches for your product on Google
• Finds up to 10 reliable sources of product information
• Automatically creates parsers for these sites using AI
• Collects data from all sources simultaneously (in parallel)
• Analyzes collected information using AI
• Creates a single, consistent product card with full specifications
• Fills out the product card in your system
• Minimal human involvement

Result: Instead of weeks of manual work, just minutes of automated processing. Complete product cards with specifications, descriptions, and images.

What problems does the system solve?

What will you get in the end?

1. Fully completed product cards — with characteristics, descriptions, images
2. High data quality — AI verifies and compares information from multiple sources
3. Save time 100 times — what used to take weeks will now take minutes
4. Automatic parsers — the system itself creates parsers for new sites
5. Detailed statistics - you can see where the data comes from and how high-quality it is
6. Scalability — processing thousands of goods without human intervention

How does it work? In simple terms

Step 1: Search for information on Google

The system generates a search query based on the product name and SKU (EAN/MPN) and searches Google. It finds up to 10 reliable sources with information about your product.

For example: If you have the product "Apple MacBook Pro 14" with the part number "MKGP3," the system will search the official Apple website, online electronics stores, and review sites for detailed specifications.

Step 2: Create a Feature Template (if needed)

If there is no characteristics template for a product category yet, the system automatically creates one using AI. This template contains all the typical characteristics for products in this category.

For example: For the "Laptops" category, the template includes: screen diagonal, processor, memory, video card, weight, operating system, etc.

Step 3: Automatic creation of parsers

For each website found, the system checks whether a parser (data collection tool) already exists. If there isn't one, the AI automatically creates a new one by analyzing the page structure.

In simple words: A parser is like a template that "knows" where on the page the product name, specifications, images, and other information are located.

Step 4: Collect data from all sources simultaneously

The system queries all sources in parallel, meaning it processes 3-10 sites simultaneously, rather than one at a time. This is 10 times faster!

From each source the following is collected:

• Name and description of the product
• Technical specifications
• Product images
• Videos and files (if any)

All collected data is temporarily stored for further processing.

Step 5: Smart Data Processing with AI

This is the most important stage! AI receives information from all sources and:

• Compares data from different sources
• Selects the most complete and accurate information
• Combines characteristics, eliminating duplicates
• Resolves contradictions (if one site says one thing and another says something else, AI chooses the correct version)
• Removes advertising phrases like "Buy now!" or "Best price!"
• Fills out your category characteristics template
• Adds missing characteristics that were not in the template
• Uses own knowledge to supplement information

Result: A single, coordinated product card with full specifications and a data quality rating from 0 to 100.

Step 6: Quality Check

The system evaluates the quality of the collected data on a scale from 0 to 100:

• 90-100 points - excellent data (many characteristics, descriptions, images)
• 70-89 points - good data (enough characteristics)
• 50-69 points - satisfactory data (basic characteristics are available)
• 0-49 points - insufficient data (too little information)

If the quality rating is below the minimum you set, the product is not enriched—the system will wait for higher-quality data.

Step 7: Saving to the product card

The system automatically:

• Adds all characteristics to the product card
• Creates new characteristics if they do not already exist in the system.
• Adds images (if you enabled this option), the source site with the largest number of images is selected
• Marks the product as "auto-enriched"
• Stores information: when enriched, from which sites the data was taken, quality assessment

Step 8: Work Log

All information about the process is saved in a log, which you can see in real time:

• What websites were used?
• How many sources were successfully processed?
• What features have been added?
• How long did it take to process?
• Errors (if any)

Description of form settings

Statistics at the top

Four counters show the overall picture:

• Total enriched — how many products have already been processed by the system
• Enriched today — how many goods were processed today
• Average quality score — average quality rating of all enriched goods
• Auto parsers — how many parsers were created automatically

Information about selected products

• Selected category — from which category the goods will be processed
• Total products — how many products match your filters
• Selected items — how many products did you check manually?

Filters for selecting products

☐ Only products without characteristics

What it does: It will only process those products that do not yet have any characteristics (completely empty cards).

When to turn on: If you want to fill only the "empty" items without touching the items that already have something filled in.

Advice: Leave it enabled when you first launch it - this way you won't waste time processing already filled out cards.

☐ Only unenriched products

What it does: Will only process items that have not yet been processed by this system.

When to turn on: To avoid reprocessing products that have already been enriched previously.

Advice: Enable this for the first run. Disable it if you want to update information in already processed cards.

System operation settings

☐ Create a characteristics template

What it does: If there is no ready-made feature template for your category, the system will automatically create one using AI.

Why is it necessary: A template is a list of typical characteristics for a category. For example, for smartphones: diagonal, camera, processor, memory. This helps the AI understand exactly what characteristics to look for and add.

Advice: Be sure to turn it on! Without a template, characteristics may be unstructured.

☐ Update existing data

What it does: If the product already has characteristics, the system will replace them with new data.

When to use:

• Included - update all information with fresh data from the Internet
• Off - add only what is missing, without touching what exists

Advice: Disable this option the first time you use it to avoid erasing any manually entered information. Enable it if you need to bulk update outdated information.

☐ Add images

What it does: The system will download product images from the found websites and add them to the card.

Important: Images will be added only if the product does not have them.

Advice: Enable this option if your products don't have images. Disable this option if you already have photos or plan to add them separately.

☐ Automatically create parsers

What it does:

• Included — the system will automatically create parsers for new sites
• Off — the system will only collect general text from pages (less accurate and large volume for AI)

When to turn off: Only for testing or if you want to save on AI queries during testing (creating a parser costs money, but saves money in the long run).

Advice: Leave it on! Automatic parsers produce much higher quality and more structured data.

☐ Debug mode

What it does: The system will save all intermediate data for verification (usually it is automatically deleted after processing).

When to turn on: Only if you need to figure out why the system collected the wrong data or why something went wrong.

Advice: Leave it off for normal use. Turn it on only if problems arise.

Numeric parameters

Minimum quality score (0-100)

What it does: If the quality rating of the collected data is below this value, it will NOT be saved to the product card.

How to choose a value:

• 50-60 - the golden mean (recommended for starting)
• 70-80 — high demands (only complete and high-quality data)
• 30-40 - soft requirements (even minimal information is retained)
• 0 - no filter (save everything found)

Advice: Start with 50 If you see that many products are being rejected, reduce it to 40. If the data is not of sufficient quality, increase it to 60-70.

Product limit (0 = all)

What it does: Limits the number of items to be processed.

Examples of use:

• 0 - process all products that match the filters
• 10 - process only the first 10 products (for testing)
• 100 - process the first hundred items

Advice: When you first start, set 10 to test and see the results. If everything is OK, put it 0 to process the entire catalog.

Number of sources for one product (1-10)

What it does: Determines how many sites will collect data for each product.

How to choose:

• 3 - fast and inexpensive (recommended for starting out)
• 5-7 — more sources = more accurate data
• 10 — maximum (the most accurate data, but longer processing time)

Advice: Start with 3 — this is usually enough. Increase it to 5-7 for rare or complex items.

List of sites to ignore

What it does: Sites from this list will be skipped when searching for information.

Format: One domain per line. For example:

amazon.com
ebay.com
aliexpress.com
yoursite.com

When to use: If you know that certain sites are providing bad data or constantly giving errors.

Advice: Add your website address to avoid loading data from your own site. Add sites after you see in the log that a specific source is performing poorly.

Step-by-step instructions for beginners

Step 1: Open the enrichment form

1. In the admin panel, go to the section "Goods"
2. Select a product category (e.g. "Laptops") - optional
3. Find and click the button Content Search > AI Automated Search
   • You can click immediately - all products in the category will be processed
   • Or first select the desired products using the checkboxes, then click the button
4. A window with an automatic enrichment form will open.

Step 2: Check what will be processed

At the top of the form, look:

• Selected category — is this the correct category?
• Total products — How many goods will be processed?
• Selected items — if you selected them using checkboxes, their number will appear here

Step 3: Set up filters (for the first time)

Recommended settings:

• ✅ Only products without characteristics - check the box
• ✅ Only unenriched goods - check the box

This means: process only those products that really need the characteristics.

Step 4: Configure the system to work

For the first launch we recommend:

• ✅ Create a characteristics template — included
• ❌ Update existing data — turned off (to avoid erasing manual work)
• ✅ Add images — included (if photos are needed)
• ✅ Automatically create parsers — included
• ✅ Debug mode — included

Step 5: Set the numbers

For first launch:

• Minimum quality score: enter 50
• Product limit: enter 1 (for the test)
• Number of sources: enter 3

List of ignored sites: leave blank

Step 6: Launch!

1. Press the big button "Start enrichment"
2. The system will start working.
3. Messages about the process will appear in the "Work Log" section at the bottom.

Step 7: Monitor the process

In the journal you will see:

• Blue messages — information on the progress of work
• Green messages - successful operations
• Red messages - errors
• Orange messages - warnings

Typical messages during processing:

1. [1/10] Item Processing: Apple MacBook Pro Laptop
2. "Starting parallel parsing for 3 sources"
3. Adding a query for: apple.com
4. "The parser is ready for apple.com."
5. "Successful parsing for apple.com - Fields: 8, Attributes: 15"
6. "Product successfully enriched. Quality score: 85.00."

Step 8: Stop (if necessary)

If you change your mind or need to interrupt:

1. Click the button "Stop"
2. The system will finish processing the current item.
3. Other items will not be processed.
4. Already processed goods will be saved

Step 9: Check the results

After completion:

1. Find the last message: "Enrichment complete. Success: X, Errors: Y"
2. Look at the counters above - they have been updated
3. Go to processed product cards
4. Check what features have been added
5. Assess the quality of the data

Step 10: Processing the entire directory

If you are satisfied with the test results:

1. In the field "Product Limit" put 0
2. Adjust if necessary Minimum Quality Score
3. Press again "Start enrichment"
4. Please wait until all items are processed.

Recommendations for settings

For beginners (first launch)

For maximum quality

To update old data

Helpful tips

1. Process by category: It's best to process one category first (for example, all laptops), check the results, and then move on to the next. This way, you'll create high-quality templates for each category.
2. Always start with the test: Process 10-20 products for the first time, review the results, adjust the settings, and only then launch bulk processing.
3. Choose the right number of sources: For simple products (a T-shirt, a pen), three sources are sufficient. For complex or rare products (specialized equipment), five to seven sources are needed.
4. Follow the quality ratings: If most items are rejected due to low ratings, either lower the minimum threshold or increase the number of sources.
5. Maintain a blacklist of websites: If you see in the log that a particular site is constantly returning errors or bad data, add it to the ignore list.
6. Run during non-working hours: If you need to process 1000+ products, run the system in the evening or on the weekend to avoid loading the server during business hours.
7. Check out the first results: After the first processing, be sure to open 3-5 product cards and check the quality of the specifications. Adjust the settings if necessary.

Frequently Asked Questions

1. How long does it take to process one item?

Answer: On average, 30-60 seconds per item. Depends on:

• Number of sources (the more, the longer)
• Website response speeds
• Is it necessary to create new parsers?

For 100 products with 3 sources it will take approximately 50-100 minutes.

2. Is it possible to stop the process and continue later?

Answer: Yes! Click "Stop," and the system will complete the current item and stop. Already processed items will be saved. The next time you run the system with the "Only unenriched items" filter, it will continue with the items that haven't been processed.

3. What should I do if the system does not find information about the product?

Possible reasons:

• The product doesn't have an EAN/MPN - please add one.
• The name is too general (for example, just "Phone") - please clarify the name
• The item is very rare or new - there is no information on the internet yet
• All found sites are blacklisted.

Solution: Improve the product title and add a SKU, then run enrichment again for this product.

4. Why is the quality rating low?

The assessment depends on:

• Number of characteristics (the more, the higher the rating)
• Availability of descriptions
• Number of images
• Availability of articles (EAN/MPN)

How to improve:

• Increase the number of sources from 3 to 5-7
• Make sure the sources you find contain complete information.
• Make sure the "Create feature template" option is enabled.

5. Is it possible to process only one specific item?

Answer: Yes! In the product list, check the box next to the desired product and click "Enrich AI." The system will process only that product.

6. Will the characteristics I filled in manually be overwritten?

Answer: Depends on the "Update existing data" setting:

• Off — your manually filled in characteristics will NOT change, the system will only add the missing ones
• Included - all characteristics will be replaced with new data from the Internet

7. Where can I see where the data came from?

Answer: The following information is stored in the card of each processed product:

• Date of enrichment
• List of websites from which data was taken

This information can be viewed in special fields on the product card.

8. How much does it cost?

Answer: The system uses AI with a reduced fee. On average, enriching one product costs 0.05 EUR (five cents). For 1,000 products, it costs approximately 50 EUR.

The exact cost depends on the number of sources and the volume of collected data.

9. What will happen to the images?

Answer: The system finds links to images and saves them to the product card. The images will be immediately available for viewing.

10. Is it possible to process products from different categories at the same time?

Answer: Yes, but it's better to process one category at a time. This will allow you to create a high-quality template of characteristics for each category.

What to do if something goes wrong

Проблема: Не удалось создать парсер для сайта

Possible reasons:

• AI не смог понять структуру сайта
• Сайт слишком сложный (много JavaScript)
• Сайт защищён от сбора данных

What to do:

1. Проверьте, что сайт вообще открывается в браузере
2. Добавьте этот сайт в список игнорирования
3. Система автоматически найдёт другие источники

Проблема: Ошибка 404, 403 или 500 от сайта

Что означают эти ошибки:

• 404 — страница товара не найдена (удалена или изменился адрес)
• 403 — сайт блокирует доступ (защита от роботов)
• 500 — временная ошибка на сайте

What to do:

• Для 404 и 403 — добавьте сайт в игнор-лист
• Для 500 — попробуйте позже, это временная проблема

Проблема: Не найдено данных на странице

Cause: Парсер не смог найти информацию на странице (возможно, изменилась структура сайта).

What to do:

1. Включите "Режим отладки"
2. Запустите обогащение снова
3. Свяжитесь с технической поддержкой, если проблема повторяется

Проблема: Процесс "завис" на одном товаре

Cause: Один из сайтов не отвечает или отвечает очень медленно.

What to do:

• Подождите 1-2 минуты — система автоматически пропустит зависший источник
• Или нажмите "Остановить" и добавьте проблемный сайт в игнор-лист

Проблема: Низкая оценка у всех товаров

Possible reasons:

• Источники содержат мало информации
• Не создаётся шаблон характеристик
• Слишком мало источников

What to do:

1. Убедитесь, что включена опция "Создавать шаблон характеристик"
2. Увеличьте количество источников с 3 до 5-7
3. Временно снизьте минимальный балл качества до 40, чтобы увидеть, какие данные собираются

Проблема: Характеристики не добавляются в карточку

Possible reasons:

• Оценка качества ниже минимальной
• Включена опция "Обновить существующие", но у товара уже есть характеристики

What to do:

1. Посмотрите в журнале оценку качества товара
2. Если оценка ниже минимальной — либо снизьте минимум, либо увеличьте количество источников
3. Проверьте настройку "Обновить существующие данные"

Conclusion

Система автоматического обогащения товарных карточек Elbuz — это ваш надёжный помощник, который освободит вас от рутинной работы и сэкономит сотни часов времени.

What you get:

• ✅ Полную автоматизацию заполнения карточек
• ✅ Высокое качество данных благодаря AI
• ✅ Быструю обработку тысяч товаров
• ✅ Детальную статистику и контроль
• ✅ Гибкие настройки под ваши задачи

Главный совет: Начните с малого. Обработайте 10 товаров для теста, посмотрите результаты, настройте систему под себя — и запускайте обработку всего каталога.

Remember: Система постоянно учится и улучшается. Чем больше товаров вы обработаете, тем качественнее будут становиться результаты, так как накапливаются шаблоны характеристик и статистика работы парсеров.

Желаем успешной работы с системой Elbuz!