PropertyGuru is a leading property management company based in Singapore, connecting property seekers to real estate agents. Their mission is to help people make confident property decisions by giving them relevant content, actionable insights and world-class service. Users of their app upload thousands of photos of their listings for rent or sale every day. In a fast-moving mobilefirst real estate market like Singapore, they needed their app experience to be responsive, accurate and be able to operate at scale at the same time. After exploring machine learning (ML) platforms and toolkits, they turned to H2O Driverless AI to implement AI for multiple use-cases.
Property Guru handles a large volume of listings and had looked to leverage AI and machine learning (ML) for multiple use-cases – image moderation, predicting churn, forecasting credit, measuring performance of listings. They realized early-on in their development that they needed machine learning techniques to manage user data, user retention and ensure the customer experience on their app lives up to their reputation. Doing this manually was not scaling so there was a real need to automate their ML process.
Solution Powered by H2O Driverless AI
PropertyGuru found that they could use Driverless AI for the entire end-to-end ML pipeline including:
- Uploading data from most of their sources into Driverless AI – images, churn, tabular data, etc.
- They could visualize this data in a few sections using the AutoViz capability and detect outliers and anomalies.
- They were able to build the model much faster using pre-existing recipes such as the churn models available. In addition, they also took advantage of the automatic model building process – feature selection, feature engineering, hyperparameter tuning and deployment.
- Lastly, they were able to seamlessly deploy multiple models directly into Amazon Web Services (AWS) Lambda service, from within Driverless AI. They were able to deploy different models simultaneously using Java objects and see their performance on live data.
With Driverless AI, Property Guru achieved the following key benefits:
- The data science team was able to iterate with new and existing models much faster than before. As they add new features into the mix, Driverless AI automatically does the feature engineering, visualization, hyperparameter tuning and, therefore, improved their agility as a team.
- Using Driverless AI enabled the non-technical teams to interact with the data more easily.
- The marketing team got a head-start with predicting customer churn rather than starting afresh with building the model.
- The data science team was also able to innovate faster and build newer capabilities, e.g. experiment with Google Lens, now that the actual model building took much less time.