Use Cases

How Hepsiburada Maximizes AI Model Performance and Search Accuracy with Co-one

Hepsiburada partnered with Co-one to evaluate search relevance, validate AI-assigned product categories, and identify duplicate listings across its marketplace.

  • Türkiye
  • E-commerce AI
  • Model evaluation
  • Search relevance
  • Product categorization
  • Product matching
  • Human review
Co-one and Hepsiburada case study cover on evaluating AI for marketplace search.

What problems was Hepsiburada trying to solve?

Hepsiburada operates an e-commerce marketplace in Türkiye with millions of products. Shoppers depend on relevant search results and accurate product categories to find what they need. Hepsiburada selected Co-one as its AI data management and model evaluation partner to address three problems:

  • Irrelevant search results. Search queries did not always match the products displayed. The team needed to distinguish exact matches from alternatives and products that complement the item a shopper searched for.
  • Incorrect product categories. An automated model might place a toy stroller under baby strollers, where it would appear in the wrong filters and be harder to find in the right category.
  • Duplicate and inconsistent listings. Multiple merchants could list the same or similar products with different information, making it harder to keep catalog records consistent.

How does Co-one evaluate Hepsiburada’s AI models?

Hepsiburada integrated Co-one’s human-based model validation into its evaluation workflows. Reviewers check search results and category predictions against defined criteria and label product pairs to identify duplicates.

Search relevance evaluation

Reviewers compare each search query with a returned product and label the result as an exact match, a complementary product, or an alternative. These judgments help Hepsiburada’s AI teams identify relevance errors and refine their search algorithms.

Evaluation volume. Approximately 120,000 data points per month.

Product category validation

Reviewers mark each AI-assigned product category as correct or incorrect. The results identify category errors that can hide products from relevant filters or place them among unrelated listings.

Evaluation volume. Between 20,000 and 70,000 data points per month, depending on volume.

Product matching and similarity labeling

Reviewers compare candidate products with a reference product to distinguish duplicate listings from similar alternatives. These labels help Hepsiburada identify inconsistent listings for the same product across merchants.

Evaluation volume. The initial project covered 200,000 data points, with additional volumes reviewed annually.

What comes next?

Hepsiburada plans to continue using Co-one’s data labeling, human review, and model evaluation expertise to monitor model performance and support new AI initiatives. The collaboration gives its teams a repeatable way to examine search relevance, category accuracy, and catalog consistency as the marketplace evolves.

More in use cases

MANK and Co-one case study cover on evaluating a banking voice assistant.

Use Cases ·

How MANK evaluated its banking voice assistant

MANK worked with Co-one to automate call-flow testing, assess how its voice assistant handled adversarial conversations, and improve speech recognition and synthesized speech.

Co-one and Akbank case study cover on evaluating a banking AI assistant.

Use Cases ·

How Co-one helped Akbank improve its AI assistant

Akbank worked with Co-one to improve how its banking AI assistant recognizes customer intent and responds. The project combined knowledge base management, automated evaluation, and human review.

Talk to Co-one

Bring a workflow. We’ll start there.

Show us where your agent makes decisions or takes action. We’ll discuss what to simulate, what to validate, and where runtime controls belong.

Book a demo

Choose whether to allow optional analytics and marketing tags.