Westpac, a $965 billion-asset bank in Sydney, Australia, has been struggling with a challenge many large banks must cope with.
Over time, the bank has developed many specialized virtual assistants that each have been trained well in a sphere of knowledge, such as bank account FAQs or helpdesk issues. But each virtual assistant or conversational chatbot is a silo with no access to information in the others. Users have to know where to go to find the answers to their questions.
To create one Google-like engine that can mine all these specialized bots, Westpac has worked with the New York technology company Kasisto to co-develop a bot that will rule all other bots, so to speak. It takes employee (and in the future, customer) questions, dives into other virtual assistants (such as within HR and helpdesk systems) to see if they have the answers, and it monitors these user conversations. If a specialized bot can’t handle a query well, the Kasisto bot will route the interaction to a human.
Building bots
Westpac has been working with chatbots for years. Since 2018 it has offered customers Red, a virtual assistant that answers basic questions and is based on IBM Watson technology.
“It's a fairly straightforward, conversational AI chatbot,” said David Walker, chief technology officer at Westpac Group. “If you want to understand what the interest rates are, Red does that reasonably well.”
It has a “containment rate” of around 70%. In other words, in about three-fourths of cases, customers’ questions are fully answered in Red and they don’t need to talk to a human. It has handled 5 million conversations since 2018. (All told, Westpac has 12.6 million customers.)
Like any chatbot, Red needs to be trained. While it has gotten smarter, it's still dealing with fairly traditional FAQs.
In 2020, Westpac launched Wendy, a “digital human” virtual assistant in avatar form that helps young people make career decisions. It’s also powered by IBM Watson’s AI conversation engine as well as technology from New Zealand-based Soul Machines, which creates avatars for the movie industry.
“We wanted to understand and explore and create an experimental mindset inside on how we could start to think about these sorts of technologies,” Walker said. “What was driving our thinking then was this idea that we're going to move from where we are today, which is digital 2.0 mobile banking and 1.0 web-based banking, and looking at digital 3.0, this more holistic view of immersion in the metaverse. No one's really doing that at scale. But if we're heading towards that sort of future, we really need to land this conversational AI piece.”
Wendy, which has a containment rate of about 76%, was Westpac’s first stab at that.
“If you put your camera on, it can monitor your facial expressions and observe how you react to Wendy, so all of a sudden you've got this incredibly powerful, two-way experience going on,” Walker said.
Wendy is meant to help people who are coming out of college and looking to start their careers. “It was a nice way to give people advice and help guide them on that path,” Walker said. “It's not always about banking; sometimes it’s about helping people in different ways. It was also a great way for us to learn and understand how and when to interact with that generation. We are now looking at Wendy and saying, What role can Wendy play across our consumer bank?”
One bot to rule them all
In addition to Red and Wendy, Westpac uses many other bots throughout the organization. There’s a chatbot within the IT helpdesk system. There are bots within ServiceNow, a workflow software program it uses in HR.
Because each specialist bot has its limitations, Walker’s team has been working with Kasisto to develop an “orchestration layer” or master bot that can pull information from any virtual assistant within which it may be lurking. Walker became familiar with Kasisto’s work when he worked at DBS in Singapore and he saw how
“If you start with a single chatbot engine, you get to this point where it runs out of power,” Walker pointed out. “Right now, even the best conversational engines in the world just can't cope with the breadth and depth of topics” an organization needs to cover.
“What we're seeing is this proliferation of chatbots that end up being silos,” Walker said. “What we're trying to do in Westpac is avoid building up multiple specialist chatbots that can each handle a subset of questions.”
This struggle to deal with specialized chatbots is common in the industry, according to Zor Gorelov, founder and CEO of Kasisto.
“That's what is happening across the board at big banks: Somebody's going and building an HR bot and somebody else is building a compliance bot,” he said. “And then there is a bot for IT support and there is a bot for consumer banking. It is confusing for customers, for employees, for everybody else.”
The master bot Kasisto built in collaboration with Westpac is a “central brain,” Gorelov said, that understands employee and customer questions and knows how to interrogate other bots. Part of Kasisto’s work was developing adapters to all of Westpac’s existing bots.
“It engages users in intelligent conversation in a single portal,” Gorelov said. “It provides a unified user interface to bots that were built by different vendors that have their own user interface standards and keywords. This orchestration layer knows how to direct the users so they don't have to learn underlying infrastructure and communication interfaces.”
Kasisto already has a conversational AI engine of its own called Kai that can also work with the new orchestration layer.
The master bot monitors every conversation, and whenever a dead end is reached it routes the conversation to a human agent.
“It's like a really smart concierge,” Walker said. “New York has some of the best hotel concierges in the world. If you want movie tickets, if you want theater tickets, if you want to ride in Central Park, they'll organize anything you want. This orchestrator, this brain if you like, is really like a super-smart concierge that knows how to get your problem solved or your task done. It receives all the requests, all the conversations coming in, whether that's through Wendy, through a chatbot on our website or our mobile banking app, or even direct SMS traffic.”
Westpac staff have been using the Kasisto orchestrator.
“If you think about the staff side of a bank, there are many places where chatbots can exist,” Walker said. “HR systems have chatbots, service desks have tickets for services. There are chatbots in different parts of the bank solving questions around legal or commercial terms and so forth.” Staff members now go to the Kasisto bot with any question. For security purposes, for some questions a user needs to be authenticated.
Eventually, the orchestrator will work with Red, Wendy and other customer-facing chatbots. All of this will help Westpac tackle the staffing shortage in Australia and elsewhere in the world.
“It just means that we can deploy our staff into areas that take them out of the mundane stuff that takes them away from helping our customers,” Walker said. “There's no shortage of needs and demands to help better service our customers. And we just want to free our staff to do even more important things, as much as we can.”
Later this year or early next year, Westpac will start piloting it with small groups of customers. The bot can handle Australian, British and American accents, Walker said. It has extra libraries loaded with Australian lingo.
The new bot will help Westpac adapt to a future in which consumers will expect ambient experiences, Walker said.
“I see my son, he watches videos, he interacts with voice. It's not a text-based world he's in; he's in a much more dynamic world of conversation,” Walker said. “Two-dimensional screens eventually will fade and this will have to be the future. So we’re getting ready for that now as it scales. And we think it's critical to set ourselves up to provide that next generation of experiences for our customers and our employees.”
