Microsoft made headlines when Tay, the chatbot designed to engage with millennials, was exploited and began tweeting what the company described as “wildly inappropriate and reprehensible words and images.”
[ Related: Microsoft ‘deeply sorry’ for Tay chatbot, will bring it back when ‘vulnerability’ is fixed ]
But other companies are quietly implementing similar artificial intelligence technology to interact with current and future customers. Atom Bank, a startup, mobile-only bank based in the U.K. recently announced that it is incorporating WDS Virtual Agent software from Xerox into its mobile app.
The machine learning software will give customers an agent-like option for assisted self-service on the app. The software was introduced two year ago to diagnose and solve customer queries by analyzing data and learning from the ways in which human agents diagnosed and solved customer problems. Over time, it develops the intelligence to understand and solve customer issues itself without having to be programmed.
Launching the first mobile-only bank, Atom Bank’s leaders had to be able to offer all the service a customer might need via a mobile app. “One of our core design principles was to enable the customer to do everything within the app,” says Stewart Bromley, Atom Ban’s chief operating officer. “This means not only that our customers can get support without leaving their banking app, but when they need support, we already know who they are, where they are, and what device they are using. Hence, the support can be more exacting to that customer in that moment.”
There were several virtual customer service agent options on the market, but Bromley says Atom Bank wanted a proven system. “It was important to us that our data and learning efforts didn’t start from square one, and WDS already has extensive experience in the mobile sector,” Bromley says. “They understand handsets, networks, devices—nobody else has this type of specialist knowledge.”
The challenges of delivering mobile customer service
While eliminating physical branches — and even Web-based options — may simplify the banking experience for customers, it creates some challenges for the bank. “Delivering mobile customer service is as complex as it gets. Not only do you need to know about your own app, which you are in control of, you have to understand the variety of devices and operating systems on which it will be used, which you are not in control of,” says Bromley. “Further, these technologies are constantly evolving and the pace of change is accelerating.”
The app, animated and built using a 3D rendering engine, is tested constantly. “We create new builds every day,” says Bromley. “Testing is crucial. Customers are very forgiving when they can see the proposition or service evolving for the better, but they do not and should not tolerate inconvenience in the process. We use gorilla-testing methods with employees, and we hold focus groups with prospective customers for each release. Once we launch, we will also invite our founding customers to actively be involved in the development of future functionality.”
The benefit of the virtual agent software is that the machine learning technology will continue to evolve in reaction to the changing environment, Bromley says.
The WDS solution offers the customer multiple potential solutions to every question. These are ranked by the probability of it fixing the issue or inquiry, based on how these answers resolved similar customer queries in the past. Each time a response is confirmed or not, the analytics are updated, improving the accuracy of responses over time.
[ Related: Consumers are fed up with ‘human-less’ customer service ]
But the system has its limitations. “I don’t think you should ever assume that the machine is always right,” says Bromley. “Further, it would be wrong to assume that the customer wants to use it. The customer will be able to use self-support, but can equally use assisted channels, such as chat, secure messaging, voice, and video. However, we also use the same machine learning behind the scenes for the assisted channels, so that all learning is applied.”