SinceraThe customer this was delivered for.Achieved 85%+ Accuracy for Millions of Records in 10,000 Categories
Key results
- above 85% categorization accuracy
- Millions of monthly records made usable
- 10,000 categories in the target taxonomy
Sincera, a company decoding digital advertising data, faced a major challenge with millions of messy, inconsistent, and unstructured product records collected monthly from various internet sources. This data was nearly impossible to use without extensive and time-consuming standardization. The goal was to map these records to Shopify’s product taxonomy, which included 10,000 categories and up to 7 levels of nested hierarchy, a feat that would be cost-prohibitive and require thousands of human hours without advanced AI. To address this, Sincera partnered with Fractional AI to build an AI categorization system using a multi-step LLM pipeline. This system evaluated each record to classify it as a brand, segment, or uncategorizable, then enriched sparse records and discerned the best matching category. The solution outputted each record to its corresponding Shopify category with a level of confidence in real-time, effectively transforming the unstructured data stream into a valuable asset. Additionally, the collaboration significantly enhanced the Sincera team's confidence and skills in AI engineering, particularly in building robust LLM evaluations. As a result, Sincera's monthly stream of messy data was successfully transformed into a valuable, usable data asset. The AI system consistently achieved categorization accuracy above 85% for millions of records. Each record was processed and categorized in real-time, providing immediate data utility and unlocking insights that were previously inaccessible due to the data's complexity and volume.
Fractional AI