About Photo Experience

We have seen from previous user research and testing that user engagement with photos is a critical component of the home buying process and lead submission. When we added a vertical photo gallery to the Android app, we saw +21% total Android leads and +14% total Android unique lead submitters. With this results, we want to take the opportunity to enhance the photo experience to be even better and solve our user pain points when they browse for homes through Realtor.com app.

My role in this project was to lead the end to end experience for iOS, Android, and Website. It included partnering with a Researcher, Product Managers, and Machine Learning Engineers.

Publications:

 

Goals

  • Help users to find photos they care about efficiently.

  • Increase in app engagement.

  • Increase in lead submissions

Problems

  • Users need to scroll or swipe through all the photos for a listing to find specific photos important to them, and this is a painful process.

  • Out of the top 25% of listings with the most photos, the average number of photos per listing is 47.

  • One listing in our app has 680 photos! Imagine scrolling through 680 photos to find only the bedroom photos.

Solutions

  • Use in-house trained ResNet50 deep learning model to categorize photos with 97% accuracy.

  • Use machine learning to breakdown photo into categories, such as Exterior, Living Room, Kitchen, Dining Room, Bedroom, Bathroom, and Other.

 

Advanced Image Tagging

 

At a high-level, all the listing images could be differentiated into two broad classes - exterior and interior. Within each of these classes, there are many tags and features. Though we could select many features and categories, some prioritized tags are essential for adding business value.

The presence of ceiling fans, chairs, tables, rugs, painting, etc. are not a significant trait in buying a home in terms of their monetary value, whereas, the presence of hardwood/carpet floors, quartz countertops, and kitchen islands will be valued in terms of consumer preference. Thus, we have carefully finalized the needed features by considering product needs, consumer preference, and domain knowledge.

 

Design

 
 

Results

Property Page

  • +8.4% Android Property page views / unique user

  • +10.2% Android List page to property page conversions / unique user

    Photo Gallery

  • +8.0% Android unique users viewing photo gallery

  • +22.0% Android taps entering photo gallery / unique user

    Photo Gallery Monetization

    • +22.5% Android leads from photo gallery in-line lead form

    • +12.6% Android leads from photo gallery bottom email lead button