case study
Boosting Real Estate Discoverability with AI
Realty Austin turned to KUNGFU.AI to boost email housing recommendations—resulting in dramatically higher click-through rates.
AI Solution(s)
Industry
Real Estate
Realty Austin turned to KUNGFU.AI to boost email housing recommendations—resulting in dramatically higher click-through rates.
Realty Austin had just revamped its website to make home buying easier, but they wanted to go further—personalized real estate recommendations delivered directly to users’ inboxes. Competing with industry giants like Zillow and Compass required something more than SEO; they needed AI to drive deeper engagement and bring users back for more.
The mission was clear: personalize communications, increase user engagement, and drive more traffic to the website. But it wasn’t without hurdles. Integrating with existing systems, selecting the right machine learning model, and deploying a scalable solution were all critical steps. Unexpected infrastructure compatibility issues added complexity, causing deployment cycles to fail unpredictably. The solution? Provision cloud resources optimized for the required CPU instruction sets.
Using clickstream data—viewed, saved, and dismissed listings—we built a recommendation engine that matched active real estate listings with each user’s unique preferences. An autoencoder model architecture allowed us to map user behavior to embedded representations of listings for real-time recommendations. To spice things up, we included “wild card” suggestions—similar listings outside a user’s typical geographic area—giving buyers a broader view and encouraging discovery beyond their usual search boundaries.
In line with our “AI for Good” mission, the model relied on clickstream data rather than user demographics to avoid potential bias in real estate recommendations, helping to combat a long-standing issue in the housing industry.
The result? Industry-leading engagement. Realty Austin’s weekly email open rate hit 32.5%, and click-through rates soared to 14.4%—far above industry benchmarks. One Realty Austin agent even noted that our AI-driven recommendations outperformed hand-selected listings by realtors, uncovering patterns and preferences that human intuition couldn’t match.
Months later, the model remains stable in production, with no major model drift and no retraining required. Realty Austin continues to benefit from a seamless, high-performing recommendation engine that keeps users coming back for more.
