Continued from Part I.
In an interview with Frontier Enterprise, Setu Chokshi, Director of Data Science at PropertyGuru Group, explains how the company uses multiple clouds for its data modelling, the role of AI and ML in enhancing customer experience, and the use of augmented reality tools that benefit both clients (agents) and their customers (buyers/tenants).
Could you talk about how large the PropertyGuru IT team is?
At Propertyguru, our teams are broken down into product lines, with the exception of the data science team. We have separate teams focusing on finance, consumer, developer and agents. All of these teams take care of different engineering aspects of the Propertyguru website itself, and each team comprises about 10-15 people. Separately, we also have a systems team that takes care of our overall infrastructure, and it is relatively smaller at about 5-6 people.
Our data science team is about 13 people strong, and we support all of Propertyguru’s data and analytical needs across the whole site.
From a cloud systems perspective, PropertyGuru’s website and most of our data science models are hosted on AWS. We’ve also started to work more extensively with Kubernetes clusters that are set up in Google Cloud Platform (GCP) for some of our data science models.
What is a typical day like for your data scientists?
We have a mix of data scientists, data engineers and data analysts in my data science team.
The data engineering team is responsible for maintaining our data warehouse, which is built around Google BigQuery and Cloud Storage for reporting, data science modeling and analytical purposes. We draw data from various systems and consolidate them into BigQuery. We use a third-party service called Fivetran to sort that data and re-enter them into BigQuery, and from there, we process that data and store it in a format that our various teams can use.
We also recently transitioned into Looker, which has essentially become our data visualisation platform, and is something that we have democratised to everyone in the organisation. So instead of always having my team be the report and content generators, we now have 6 of the top 10 platform users who do not actually come from the data science team. Looker has hence enabled them to become the content generators, allowing them to become part of that process themselves.
My analytical team provides more support for businesses, helping them to make the right set of decisions and to work through more complex analytics that are not possible using regular approaches. For this, they partner with our data scientists to utilise various data science techniques to generate viable solutions.
The data science team itself spends a majority of the time working through different data sets, getting them ready for our various modelling purposes. The data science team also partners with the product team to work on any use cases, along with any purpose-built products we want to deploy, ensuring that we are able to generate value for our consumers, agents and developers on the PropertyGuru platform.
Could you talk about how PropertyGuru is using AI & ML?
For us, AI plays a very important role in the business itself, and it goes hand in hand with our vision and mission at PropertyGuru. Our vision is that we want to be a trusted advisor for anyone who is seeking property. And the way we are going about achieving this – our mission – is to make confident property decisions, through relevant content, actionable insights and world-class service. Both relevant content and actionable insights are what my team supports and works actively towards.
Until recently, agents will input their phone numbers, an image of their faces, and their call-to-action notes on a property photo. But since April of last year, we have enforced that these should not be on the site as they do not translate into a very good user experience for consumers. Our Computer Vision tool enforces this, helping us to ensure our agents are not trying to bypass the system or create a distracting experience for consumers.
In the same spirit, we also launched PropertyGuru Lens last year, which allows consumers to search for properties on the fly. So if you’re in the locality, you can whip out the PropertyGuru app, use the camera function and point it towards a building, and the app will let you know the building you’re looking at, along with how many units are on sale or for rent. This helps to create a seamless experience where consumers can easily point and click, and see what kind of properties we have on our site.
We also take steps to ensure that our platform is free from bias.
Some agents or landlords may not wish to lease properties to certain communities, and input criteria such as “no Indians”, “no PRCs”, and so on in their property listings. Such language is not welcome on our platform, and we use AI to actively police for such listings and reach out to advise agents that these are not acceptable.
Recommendation systems also play a very important role on our platform. After a consumer interacts with two or three clicks on a particular property, the recommendation systems will tune in and provide recommendations on the types of properties they should look at, based on the kind of properties they have been searching for and leads they have been sharing. This eliminates the need for consumers to explicitly ask us for any specific kind of information, and hence also covers a big part in improving our consumer journey.
Beyond customer experience, PropertyGuru also ensures that our agents are given a good perspective of the property market. We provide our agents with as real-time information as possible – not just on their properties themselves, but on general market insights as well, including district-level insights, and even micro-neighbourhood insights. This allows our agents to better serve their clients, providing relevant inputs and advice as needed.
These are just some of the key products that run on AI and ML that we use very actively to enhance user experiences on PropertyGuru’s platform.
What technology is being used to ensure that customer experience is smooth, how is PropertyGuru upgrading its infrastructure to ensure that it’s ready to scale up?
Kubernetes is something that we use very actively in PropertyGuru, and it allows us to effectively scale our hardware resources on the cloud to ensure that it makes for the same user experience for our consumers even with high volumes of traffic to the platform.
Besides this, we are utilising a lot of AWS technologies to deploy our data science products and scale up and down depending on workloads. That being said, we are slowly migrating a lot of that workload into Google Cloud, just because of the inherent nature of how well Kubernetes is supported on the cloud itself.
For instance, Google Maps is one of the platforms PropertyGuru uses extensively.
We leverage a lot of deep APIs from Google Maps, be it points of interest, the map locations themselves or location insights.
These are powered on PropertyGuru using Google Maps itself.
Subsequent to that, we also have some of our workloads, specifically our recommendation systems, which are working on a Kubernetes cluster in Google Cloud.
How do you see the online real estate industry evolve? What are some of the features that PropertyGuru is working on at the moment to further enhance its portal?
Our consumers are the ones who drive our business decisions, and we always strive to go to where our consumers want us to go. In light of COVID-19, we fast tracked a new digital feature in our property sales enablement platform FastKey. Called FastKey Storyteller, it brings fully digitized, 360 immersive walk-throughs of developer projects, Amenities, Units and surrounding cityscapes right at buyer’s fingertips. With this new feature, FastKey now enables property developers to go-to-market as soon as their project is approved and without waiting for the construction of the physical sales galleries or showflats. This unlocks project discovery and immersion from across the region, which is timely as we enter a new normal of social distancing and travel restrictions.
We are also piloting a lot of other products internally, and we conduct focus group tests before we launch them, which is an ongoing process. Based on what the consumers, agents and developers are looking for, we will be launching the appropriate products to help address their needs.
