BUILDING GENERATIVE AI PERSONALIZED RECOMMENDATIONS
Factored partnered with one of the largest retail chains in the United States to build a gen AI tool that could understand the needs of any customer and make personalized recommendations as easy as having a personal shopper. This would represent a critical competitive edge as according to a 2024 Deloitte study, 80% of consumers surveyed prefer brands that offer personalized experiences and reported spending 50% more with such brands.
Fixing our data gaps and infrastructure
The retailer’s existing recommendation system was ineffective. It had access to less than half of all common product information and data on just a small fraction of alternative products. The model also lacked the context of a shopper’s cart, purchase history and preferences. This resulted in poor personalization and missed opportunities to influence behavior.
Factored integrated with the team
Factored joined the retailer’s team backed by our Centers of Excellence in Data Engineering, Data Analytics, Machine Learning and Software Engineering.
We designed our solution with efficiency in mind
We chose a leading LLM with a robust knowledge set and a high quality of response. Additionally, we chose RAG to contextualize the LLM into the retailer’s data. To preserve data and privacy, operate without restrictions and reduce costs, our solution also used a local LLM.
Our RAG Framework was broken down into 2 workflows:
I. Preparing the data required to give the LLM context for the prompts and understanding the user’s intent.
II. Delivering meaningful recommendations.
Components
• Data warehouses.
• Tabular data to process and generate context for the prompt and our reply.
• An embedding model to convert sentences and paragraphs into a fixed-length numeric representation.
• Hybrid similarity search endpoint to retrieve the context from the vector database.
Delivering expert results
We evaluated our RAG pipelines with a method that uses other LLM as judges.
Threshold >=4
GROUNDEDNESS
Can the questions be answered from the given context?
4 out of 5
RELEVANCE
Is the question relevant to users?
4 out of 5
STANDALONE
Is the question understandable free from any context?
5 out of 5
FAITHFULNESS
How factually accurate is the generated answer?
4.5 out of 5
Result
QUERY
I want a new pair of jeans that are a little stretchy and environmentally friendly.
RECOMMENDATION:
Hello there! I’m happy to help, here is one of my top three recommendations available in your size:
Diesel straight jeans 1985 larkee 009zs:
This dark blue, straight-leg jean features a relaxed waist. Made from 95% cotton, 3% elastomultiester, and 2% elastane -spandex, it offers stretch. The Jacron patch is crafted from FSC® certified cellulose fiber, sourced from responsibly managed forests.
