Using an AI Agent to give Beauty Product Recommendations

Problem

A big retail company, one of the largest in Europe, offers an extensive range of products on their online store. Traditionally, customers had to select specific keywords to find the products they wanted.

Now imagine an assistant that knows all the available products in the shop, interacts with the customers in a human-like way, comprehending their needs and ultimately recommending the most suitable products. Customers no longer need to search for products with specific keywords. This removes any guesswork, making the shopping experience smoother, more intuitive, and personalized to their preferences.

Solution

Working with the client, we developed a shop assistant—a Large Language Model (LLM) based agent designed to interact with potential customers through a chat widget. This assistant understands their needs and uses web services (like search engines) to find and recommend the best products from the online shop.

The agent engages in fluent conversations on any beauty-related topic, adapting to various inquiries while ensuring responses are safe and unbiased. In addition to using LLM capabilities to understand topic and intentions, several other mechanisms were explored, including classification and zero-shot.

All the interactions are stored, automatically analyzed, and monitored to continuously refine and enhance the agent’s performance, using an external LLM evaluator.

Impact

The impact of the project has exceeded the high expectations. Initially designed as a prototype for a specific section of the online shop, the client is now planning to expand it to other retail product categories.

The successful execution of the project has dispelled doubts and alleviated many fears, paving the way for new AI initiatives.