Looking to buy a home? Here’s how an AI chatbot can help

5 min read

Looking to buy a home? Here’s how an AI chatbot can help

As developers at UBank, Australia’s leading digital bank, we’re constantly reconfiguring the way we work to produce the next big innovation for both our business and the digital banking industry. Disruption through innovation is at the heart of everything we do. One of our most recent innovations is RoboChat, Australia’s first AI chatbot to help customers through the online home loan application process.

How did this idea come about? After working with the IBM Cloud Garage team, we knew we wanted to build a cloud native chatbot utilizing AI. Our leadership team had been looking at providers of cloud based cognitive services for the past few years, and IBM Cloud stood out because of its Watson services and their relevance to what we wanted to achieve.

Scoping and developing the AI chatbot

We specifically wanted to use IBM Watson services to improve to improve our customers’ experience when they start a home loan application with UBank. Consisting of only twenty-eight questions, our home loan application short form usually takes around four minutes to complete. A LiveChat option, using the LivePerson platform, is on the application page and enables people to contact a UBank Advisor if they need any help. We knew that a lot of people were using LiveChat for answers to relatively simple questions, such as “What is the current interest rate?” With this in mind, a great opportunity presented itself to handle these questions through automation, using IBM Watson, which would then free up our Advisors to help customers with more complex questions.

With this goal, we began building RoboChat.

Using thousands of questions pulled from previous chat transcripts, RoboChat was built and trained with IBM Watson Assistant in just eight weeks.  The application logic built with Watson Assistant involved identifying main patterns in our transcripts, and defining short and long tail answers. Using Natural Language Understanding (NLU) components of Watson Assistant, RoboChat answers users’ question by resolving a relevant set of entities and an overall intent, providing the most appropriate response straight away.


Customising the AI chatbot

By being the first “Advisor” the customer encounters, RoboChat integrates with the LivePerson chat platform that remains in use with the online home loan application form. Depending on Watson’s confidence level, RoboChat attempts to answer customers’ questions or requests clarification. If RoboChat still lacks a confident answer after three consecutive failed attempts, it automatically transfers the chat session to a human Advisor. The customer can also request to be transferred to a human Advisor at any time during the interaction.


To ensure a smooth switchover to our Advisors, the UBank digital team created a software adapter which orchestrates communications between Watson components and the LivePerson chat platform. Integrating with Watson was straightforward, as all required services are hosted on IBM Cloud: A PostGres SQL database, a Node.js application runtime, as well as ancillary monitoring and notification services. Connecting these services together literally took a few clicks in the IBM Cloud console. On the live chat side of things, with assistance from LivePerson’s engineering team, we integrated Watson with the chat platform using its native RESTful APIs, using Watson Developer Cloud SDK.

We also created a fully automated pipeline for the testing and deployment of RoboChat’s orchestrator, using the built-in and open IBM Cloud toolchain. Thus, every code change committed into the application’s Git repository kicks-off a build that is tested and, if ready, deployed into a load-balanced, auto-scaling environment running Node.js runtimes. The process is completely hands off, allowing for true continuous integration and deployment (CI/CD) in both development and production environments.  This means we are spending less time supporting the application and more time working on improving the code and adding new features. For a deeper dive into microservices architectures, view this guide.

The future of our AI chatbot

While the training of RoboChat is still ongoing, with the transcripts from conversations being iteratively used to help improve the confidence level of its answers, it currently operates at a confidence level of over eighty-five percent. That reflects a twenty percent increase in confidence level since the early launch days. As RoboChat learns, and its confidence level steadily improves, we’re seeing less and less of its conversations being transferred to human Advisors. Based on our initial goal of developing RoboChat to help with the home loan application process and free up our Advisors to handle more complex questions, we consider the project wildly successful.

So, what’s our next goal? We want to reduce manual intervention required from our developers by further training RoboChat through true machine learning capabilities. Stay tuned!

Want to build a chatbot yourself? Start your team’s next project free with IBM Cloud.

Be the first to hear about news, product updates, and innovation from IBM Cloud