Case Study 04
Disrupting Real Estate with AI.
Finding the perfect home is about more than just price and square footage—it’s about lifestyle, ambiance, and personal preferences that traditional real estate search tools often overlook. At Flyhomes, in November of 2022, I led the research and contributed to the implementation of an AI-powered search feature designed to bridge this gap.
By leveraging natural language processing and a personalized “match score” system, we transformed unstructured listing descriptions into structured, searchable data. Over the course of four months, our solution, as featured in Business Wire and USA Today, made home searching more intuitive, leading to a 46% increase in tour requests and nearly doubling our sign-up conversion rate.
This case study explores how AI-driven insights helped users find homes that truly fit their needs—faster and more efficiently.
The problem.
Most real estate websites prioritize structured data—price, beds, baths—while overlooking personal preferences like natural light, architectural style, or proximity to grocery stores. This forced users to read through the property detail paragraph manually, making home-searching tedious and frustrating, like trying to spot the perfect tree in a dense forest. Despite a surge in traffic from our paid social campaigns, only a fraction were successfully finding homes they loved or booking tours—an essential step in engaging with Flyhomes.
The Solution.
Personalizing the Experience
Users are able to set up detailed preferences by selecting home feature tags and decide on level of importance.
Saving Homebuyers Time
The listings are prioritized by match scores based on the features of each home, which reduces the number of listings users have to go through.
Making the Process Efficient
The home feature tags are presented on the listing detail page, allowing users to digest the information more efficiently.
The Impact.
83%
Increase in feature completion
46%
Increase in tour rate
44%
Increase in sign-ups
Approach and Timeline.
Phase 01: Discovery & Design: Nov 2022 (3 months)
We designed an AI-based home search feature by first conducting user interviews to identify missing yet important search attributes, such as neighborhood ambiance and natural light. Analyzing over 10,000 property listings helped us prioritize these insights in a custom questionnaire.
Phase 02: Iteration and Launch: April 2023 (2 months)
We ran usability tests that led to improved tooltips for real estate terms, conducted card sorts to refine the information architecture, and validated the new search flow through A/B testing, which showed improved user satisfaction and increased tour requests.
Research: Phase 01
User Interviews.
We interviewed prospective homebuyers and current customers to understand which attributes mattered most but were missing from typical search filters.
Participants frequently cited lifestyle-related factors like proximity to grocery stores, ambiance of a neighborhood, or the “feel” of natural light in a home.
Data Analysis.
We then conducted an AI-powered analysis of over 10,000 property listings, focusing on unstructured data fields (agent remarks).
This helped us map out how often key attributes appeared, so we could design our questionnaire around fields that were both high-value to users and frequently mentioned in listings.
Questionnaire Design.
Based on these findings, we created a questionnaire flow that captured users’ unique preferences.
We included high-coverage attributes (e.g., “close to schools,” “modern style”) and some lower-coverage but high-value ones (e.g., “lots of natural light,” “nearby restaurants”) to account for various user needs.
Journey Mapping the Experience.
Hypothesis.
We believed that systematically capturing user preferences and converting unstructured listing descriptions into structured data would lead to:
- Higher Engagement – Users would be more engaged when searching because relevant results surfaced faster.
- Increased Tour Requests – With a tailored set of listings, users would be more inclined to schedule tours.
- Better Lead Generation – An interactive and personalized quiz would draw in potential buyers, improving the conversion rate from site visitor to lead.
AI-Processed Listings.
We utilized ChatGPT + ScaleAI to parse agent remarks for hidden attributes.
A dedicated back-end service tagged each listing with structured fields like “lots of natural light” or “big backyard.”
Match Score Algorithm.
Each user’s questionnaire responses were translated into a preference profile.
Our system calculated a personalized “match score” for every listed property, sorting results to highlight top matches without entirely excluding lower-scoring ones, mitigating the “fear of missing out.”
Research: Phase 02
Usability Tests.
we ran three rounds of usability testing with 5 users per round, iterating based on feedback. This led to the addition of helpful tooltips for real estate terms like “open concept,” improving clarity.
Card Sorts.
I conducted card sorts with 10 homebuyers to inform the platform’s information architecture, ensuring intuitive content grouping.
A/B Testing.
A/B tests compared the new AI-based search against the previous experience.
Success metrics like completion rates, user satisfaction, and in-person tour requests confirmed the new feature’s effectiveness.
Key Insights
Insight: Participants thought of the home tags as filters, due to their established mental model, and were concerned that they would see less listings.
Insight: Participants were factoring both must-have and nice-to-have features when browsing listings.
Iterative Refinements.
Multiple rounds of design reviews and user testing informed adjustments: simplifying question wording, adding relevant attributes, and refining how results were displayed to avoid overwhelming users.
- We adjusted weighting in the match score based on user feedback—some discovered “nice-to-have” features shouldn’t overshadow fundamental preferences like location or price range.
- We compared the AI-Based Search flow against the existing search experience. Key performance metrics included completion rates, user satisfaction scores, and percentage of users who requested an in-person tour.
- We redesigned the flow to provide a weight to each response, letting them express the importance of a feature, rather than just turning it on or off.
Final testing showed strong positive responses, with user feedback suggesting it enabled them to save a significant amount of time, which is exactly what we had hoped for.
How the solutions work together.
The feature utilizes data extracted from listing details and user-defined home features to calculate a match score for each listing. As a result, listings are ranked and prioritized based on this score, providing more personalized and smarter recommendations, and ultimately helping users find their ideal homes more efficiently.