By Jeremy Hubbard, Head of Digital & Technology, UBank
In May, we launched RoboChat, the first chat bot in Australia to help customers with their home loan applications. While we’ve only been live for a few weeks, we’ve already learnt a great deal.
Although our collective ability to apply the power of Artificial Intelligence (AI) is still in its infancy, the technology is already infiltrating all of our lives. How we work, entertain ourselves, the music we listen to, even how we make basic, human decisions, are now all open doors for AI. For our team, the opportunity was to leverage AI to create a more personal experience for our customers. Both in terms of the content we give them and being available whenever and wherever our customers want us. Having now dipped our toes into the AI pool with RoboChat, we’re continuing to learn with the ultimate goal of delivering a truly one-on-one experience.
Our journey to launching RoboChat started with a clear goal – to help make the home loan application process simpler, better, and smarter. We reviewed a variety of concepts with our partner IBM and narrowed in on a chatbot as the most direct way to help our customers fill out the application form.
Reviewing all of the home loan data we had on hand, we uncovered a key insight.
One of the key learnings is that managing context is challenging for AI in general
Approximately, 80 percent of our customers ask the same or a very similar set of questions as they fill out their home loan application forms.
For example, they want to know, “What’s a variable rate?” or, “What term do you offer?” and they want simple answers. It was clear to us that this is where we needed to focus RoboChat.
Our analysis then distilled this 80 percent of asks into 40 core topic areas. And from there, we identified thousands of associated questions across those 40 topics and trained RoboChat in how to respond.
Like any other UBank advisor, RoboChat was trained in our brand tone of voice, taught to avoid jargon at all costs, and only offer customers the information they need, and nothing they don’t. The final piece in bringing RoboChat to life was the technology integration.
We built an orchestration layer to connect IBM Watson’s Conversation API with our LiveChat solution, LiveEngage by LivePerson. The build went smoothly and integrated RoboChat right into our existing LiveChat interface.
Our learnings so far
We’ve only been live for six weeks, but we’ve learned a lot already. One of the key learnings is that managing context is challenging. While it’s easy for us humans to refer back to what we spoke about five minutes ago and link a new question to that prior conversation, this is a real challenge for RoboChat and something we’re still working on.
We’ve also seen a lot of positive feedback from customers who have used RoboChat to help with 4-5 questions and then carried on completing the home loan application. This is exactly the type of real-time support we were hoping to provide, so we’ll continue to learn and make the experience even better.
First on the list is tapping into the tone analyser offered by IBM Watson to allow RoboChat to understand and respond based on customer sentiment. While we’ve barely scratched the surface with AI, we’ve certainly taken a big leap. We’re iteratively building RoboChat’s competency, and learning with the technology as we go. As our environment continues to evolve, we will continue to be a bank of ‘firsts.’ The way customers manage their finances is continuously changing, and we’re right there with them.