Lowe's, the Fortune 50 home improvement retailer, has significantly improved its e-commerce data quality and search accuracy by leveraging OpenAI's fine-tuned language models, resulting in a better shopping experience for customers.
Lowe's, processing approximately 16 million customer transactions weekly in the United States, has taken a innovative step to enhance its online shopping platform. The company's Data and AI team has successfully implemented OpenAI's GPT-3.5 model to address long-standing challenges in product data accuracy and search relevance on Lowes.com.
The initiative began in January 2023 when Lowe's engineers and AI specialists started using OpenAI's GPT-3.5 model to tackle data quality discrepancies. Initial tests showed promising results in improving accuracy, prompting the team to explore further applications of the technology.
One of the primary challenges Lowe's faced was ensuring accurate product descriptions and search results. For instance, customers searching for "copper bathroom faucets" or "18-inch dishwashers" would often encounter irrelevant results due to inaccurate product catalogue titles or descriptions. This issue not only created friction in the online shopping experience but also made inventory planning and organisation difficult.
Lowe's team employed prompt engineering techniques, experimenting with various combinations to enhance the accuracy of the AI-based product classification system. Within months, they observed significant improvements in specific product categories and scaled the process further.
The implementation of this AI-powered solution immediately reduced the workload of associates responsible for vetting errors in product descriptions. However, to further improve accuracy and efficiency, Lowe's utilised OpenAI's fine-tuning API to customise the GPT-3.5 model for better understanding of Lowe's specific dataset. This fine-tuning process led to a remarkable 20% increase in accuracy.
The results have been substantial. In some product categories, Lowe's rate of detecting errors has increased by approximately 60%, leading to significantly more accurate product descriptions in search results. This improvement has had a direct impact on the customer experience, reducing friction across the shopping journey and driving e-commerce growth.
Looking ahead, Lowe's plans to continue refining its approach by fine-tuning newer versions of the model based on errors flagged by team members. This ongoing process aims to further minimise errors and increase productivity in the error vetting process.