11,720 product fields in 4 days: how LabelFort structured 1,465 e-commerce URLs at scale

The client builds e-commerce intelligence and product understanding systems that depend on normalized product attributes - not raw page HTML. Their pipelines need structured price, rating, category, and descriptive fields extracted consistently from live product URLs so downstream models can monitor catalogs, compare pricing, and power recommendation systems.

LabelFort processed 1,465 e-commerce product URLs and delivered 11,720 validated labeled fields within 4 days, using a maker-checker workflow across 14 annotators with field-level QA on every record.

The challenge

E-commerce product pages vary widely in layout, formatting, and category-specific information blocks. Even standard fields like rating and price appear in different positions, with different formatting symbols, and across seller-specific page designs.

The client needed structured extraction at scale - with strict rules so price, discount, and rating fields were not confused across annotators.

• Variation in product page formats across categories and sellers
• Multiple price representations - actual price, discounted price, deals, and coupons
• Correct extraction of discount percentage without field confusion
• Accurate rating metrics - overall rating and rating count from verified sections
• Clean capture of About Product content with consistent formatting
• High accuracy across large volumes through multi-person execution

What we delivered

Each product URL was processed and labeled across eight structured fields:

1. Product ID
2. Product Category
3. Actual Price
4. Discounted Price
5. Discount Percentage
6. Product Rating (overall)
7. Product Rating Count
8. About Product

This framework enabled both quantitative and descriptive understanding of each product record, giving downstream systems normalized attributes rather than unstructured page text.

Our approach

LabelFort ran the engagement as a maker-checker workflow: annotators extracted fields from each live URL under shared standards, and QA reviewers validated every output before delivery.

• Makers identified product ID and category from page context and captured prices exactly as displayed.
• Rating and count metrics were extracted from verified rating sections only.
• Checkers confirmed field-level accuracy and prevented actual vs discounted price confusion.
• About Product sections were collected using consistent formatting standards across all records.

Results

The engagement delivered a structured, standardized product dataset on schedule - every field validated through maker-checker QA before it shipped.

• 1,465 e-commerce product URLs processed
• 11,720 structured fields extracted and validated
• 78 hours 25 minutes of total agent effort
• 14 annotators on a maker-checker workflow
• Eight fields per URL - ID, category, pricing, ratings, and About Product

Key insights

• Maker-checker review improves extraction quality when web page structures vary.
• Price extraction requires careful differentiation of deal price, MRP, and coupon formats.
• Ratings and counts must be extracted from consistent source sections to avoid mismatches.
• Standardizing About Product text improves downstream product attribute understanding and classification models.

Impact

• Price monitoring and competitive catalog analytics
• Standardized product attribute pipelines for e-commerce intelligence
• Recommendation and product understanding systems
• Training data for product page parsing and attribute extraction models

Key takeaways

• Extracted 11,720 validated fields from 1,465 e-commerce URLs in 4 days through maker-checker QA.
• Eight-field framework normalized quantitative and descriptive product attributes for downstream pipelines.
• Checker validation prevented price field confusion and rating source mismatches across variable page layouts.
• Consistent About Product formatting kept text-rich fields model-ready across categories and sellers.