Real Estate Brokerage Offers a Uniquely Delightful Search

A premier real estate brokerage partnered with Hint Services to revolutionize their property search experience, leveraging natural language processing to create a more intuitive and powerful search capability that better serves their customers’ needs.

The Challenge

The draw toward natural language for product search has become evident after the introduction of ChatGPT and ChatGPT Plugins. Teams that traditionally relied on application design, information architecture, and SEO schemas have new competition from conversational interfaces and modern natural language querying techniques. To stay relevant, organizations need to match customer expectations to receive relevant search results based on a wider set of criteria than ever before.

Our Approach

Working with Hint, the real estate brokerage was able to improve their search capabilities to complement their optimized traditional filtered search experience. Our team helped to level up their developers with knowledge about vector stores and embeddings models and the architectural patterns that enabled the implementation. We used our vector test suite, Vecta Testa, to help select models that fit their expectations.

The new feature set allowed customers to find listings that were otherwise hidden due to the rigidity of query language filtering and the clumsiness of existing natural language search solutions. By tying search optimizations more closely to their data, the team was able to deliver high quality search results to their customers.

Ideation & Strategy

E-commerce experiences designed to allow customers to shop products often miss the mark because neither the products nor the customers are involved in the development of the underlying systems. Because products change over years, teams designing these experiences opt toward generic interfaces that don’t uniquely highlight products for the customer. The underlying systems similarly serve generic keyword search capabilities aimed at serving any product using a few search parameters. This genetic design leads to a bland application experience and an incomplete search experience.

Natural language search and chat enables the interface to better serve customers and products. By directly responding to queries with information about the product, we can specifically highlight aspects that matter to customers and that are unique to our product offerings. This better serves customers and allows for a significant improvement in search flexibility.

Implementation Details

Teams looking to improve search are heavily tied to the data they have available to power that search. Our implementation focused on several key areas:

1. Data Aggregation and Cleaning

   • Aggregated relevant property details
   • Cleaned out irrelevant or missing fields
   • De-cluttered data to eliminate erroneous reports
   • Enabled more consistent testing experience

2. Model Selection and Testing

   • Tested a range of scenarios and cases
   • Selected appropriate product attributes
   • Improved criteria for model selection
   • Evaluated multiple embedding models beyond OpenAI

3. Results Processing

   • Implemented secondary sort functions
   • Achieved more relevant product ordering
   • Outperformed compared service offerings
   • Maintained flexibility for future improvements

4. Future-Proofing

   • Codified results in test suite
   • Enabled future experimentation
   • Created flexible architecture
   • Allowed for adoption of new technologies

Customer Impact

With these enhancements made possible by natural language processing and machine learning techniques, customers experienced significantly improved search capabilities on the website. They could now perform very specific searches based on multiple criteria rather than being limited by basic keyword searches. The brokerage found that this greatly increased customer satisfaction and engagement with their platform.

Looking Ahead

The implementation of this advanced search capability has opened up new possibilities for the real estate brokerage. Future enhancements may include:

1. Integration with conversational AI for more natural dialogue-based searches
2. Enhanced personalization based on user preferences and search patterns
3. Expansion to additional property types and markets
4. Integration with virtual touring and visualization technologies