What digital humans can teach us
- 22 April, 2019 06:30
Due to our digitally connected world, people expect real-time answers to their questions
Would you actually ask your bank officer about the weather? To tell you a joke? How to go home early? Or even holler, “My ***** house is on fire?”
Yet, these are just some of the situations Jamie, an employee at ANZ Bank in New Zealand, finds herself in everyday, at all hours of the year.
Jamie is ANZ’s digital assistant, who helps customers with general banking questions.
Jamie has a human face, voice, expressions, and persona, and is brought to life using Soul Machines’ Human Computing Engine (HCE) – a virtual nervous system modelled on the human brain and nervous system, says Liz Maguire, head of digital and transformation at ANZ Bank.
Jamie has been on the job for less than 10 months and can provide answers to common questions, including how to open bank accounts and send money overseas.
Jamie, however, does not ask for personal information or give financial advice.
Maguire, nonetheless, points out that customers come to new technologies expecting they know what a digital service looks like.
“They have a bunch of expectations and are expecting quite a big degree of capability,” says Maguire, who spoke at a recent Trans–Tasman Business Circle luncheon in Auckland.
First, they expect real-time answers.
“Due to our digitally connected world, people expect real-time answers to their questions, and Jamie is no exception,” she says.
Jamie is a digital channel with a human touch
For ANZ, the biggest challenge is programming conversational content within Jamie.
“The technology was relatively easy,” says Maguire. “Don’t underestimate the work in content.”
“You can’t just get website content and turn into a conversation.”
Jamie’s responses need to make sense in a conversation, in a way that fits her persona as a 20-something adult.
“We know that artificial intelligence must be able to recognise the theme of the question, and that question could be asked in one of many ways,” she explains.
“We need to be able to accurately work out the sentiment analysis for those kinds of conversations.”
Thus, when the team is programming her content, they always ask, “What would Jamie say?”
“It is about establishing a back story,” states Maguire.
ANZ views Jamie as a digital channel with a human touch, she adds.
To humanise Jamie further, she discloses that they hired a screenwriter to develop Jamie’s personality and interests outside of banking.
Thus, Jamie can tell some ‘Dad jokes’ or talk about what movies she likes. She will tell you her favourite colour is ANZ blue and her favourite TV programme is Country Calendar.
At the same time, they never pretend that Jamie is human.
“She very clearly is not human,” Maguire stresses and adds that if you ask where she lives, she would say, “the cloud.”
“Customers often confuse digital channels with being quite human,” she states.
For instance, when the bank changed the voice they used in telephone banking, a customer rang them to say “The man talks too quickly”.
“Customers quite often think there is more humanity involved in the channel,” she explains.
Maguire says there is one constant in working with Jamie.
“It would be imprudent to crowdsource her learning,” says Maguire.
We know people will be more abusive over the phone or online, than when they are in person
“She is totally moderated, there is no chance of her going off script and learning some bad things.”
She points to Microsoft’s experience with its chatbot ‘Tay’, that “was very sweet and within 24 hours became horribly sexist and racist.”
Maguire relates that when Jamie gets abusive comments, she will respond, “‘Don’t talk to me that way.”
When she has worked out what your intent is, she will stop smiling and suggest that you call a number and will then close the conversation.
This is not surprising, Maguire states.
“We know people will be more abusive over the phone or online, than when they are in person."
Interestingly, before they launched Jamie, their research indicated customers did not feel comfortable using a digital assistant who knew their name and personal information.
User behaviour once Jamie went live, however, shows otherwise.
Maguire shares that customers ask Jamie for their personal account balances and other actions on their behalf, such as getting real-time rate information, retrieving their passwords or their customer number.
“User behaviour shows there is demand for this kind of technology and these capabilities are all on the roadmap for Jamie’s development,” says Maguire.
She adds that they have updated Jamie so she can provide customers with real-time mortgage rates and information on planned digital outages.
Maguire says over the past 10 months, Jamie has had more than 21,000 conversations with existing and potential customers.
The most common question asked her is how to open a bank account (asked over 2000 times).
More than 500 people have asked Jamie to tell a joke. One of her jokes runs like this: “A Freedom Account asked a Jumpstart Overdraft out on a date. Jumpstart said no, he had no interest.”
More than 800 customers have asked Jamie for weather updates from their region.
“We get lots of weird questions,” says Maguire, smiling.
These include: “How do I open an illegal bank account in the Cayman Islands?” “Can I get a million dollars in my credit card” and “How do I get to go home early?”
Maguire says there are three questions people ask when interacting with digital tools: “Is it safe? Is it relevant? Is it credible?”
It is about establishing trust, she explains, and having a technology that is going to solve an extra problem or improve one’s life.
She further adds that having a digital assistant like Jamie is about providing customers with a choice on how they would like to interact with the bank.
“We understand that customers can already do a lot of their everyday banking online but there are certain things people want to speak to humans about and we do not see this changing.”
She also says, “Jamie provides customers a unique, digital way of finding information through conversation. ANZ staff have responded positively to Jamie and we still have bankers in branches at the end of the phone and available online.”
When Jamie was first introduced, she could answer questions across 30-banking related topics based on the most frequently asked questions in the help section of the ANZ website.
“As customers used her technology and we built an understanding of the kinds of questions she was asked, we programmed Jamie with a greater knowledge base across topics people are asking her about, “ Maguire explains
Six months into her pilot phase, she could successfully answer around 200 customer questions across a wide range of topics, including more playful questions such as ‘What’s your favourite movie?’, says Maguire.
She says Jamie officially moved out of the pilot phase at the end of March this year, and they are no longer assessing her effectiveness on the number of questions she can answer.
“Instead, we’re looking at how well she is answering the questions customers are asking, taking into consideration her technical limitations,” shares Maguire.
Maguire says they have learned a lot from this “exciting journey” of creating a digital human.
She recalls a conversation with Adam Cutler, distinguished designer At IBM.
We get lots of weird questions
His view, though, is for AI to demonstrate authenticity, “a kind of vulnerability”, she says.
Maguire says they are looking at how Jamie can be “more accessible” when answering customer queries.
According to Maguire, Jamie is currently very transactional when responding to questions.
She cannot follow a conversation with a range of topics and cannot remember what the customer asked two minutes or two days ago.
“It is hard for her [to do this] if she cannot remember what you said at the beginning,” she explains.
“We continue to explore how to best provide Jamie with a ‘memory’ and better create continuity between an interaction with Jamie and a handover to a staff member, when required or preferred.”
Maguire would also like to see Jamie be able to operate inside ANZ’s other digital channels, such as internet banking and goMoney, and perform tasks for other people.
One such task could be transferring money from their everyday account to their credit card.
Again, this requires some consideration for what customers want.
Maguire says goMoney has almost a million active users. “You will do it in a way that is not annoying. Like when you are checking your account balance and Jamie is popping out [on screen].”
She also says they are also looking at changing Jamie’s avatar over time.
The current avatar is a stock one based on an employee at Soul Machines.
She shares that a key factor in naming the avatar ‘Jamie’ is that it is gender neutral.
“Over time, we will definitely change and people will be able to choose what kind of avatar they want to deal with.”
Deploying digital humans? The tech is relatively easy, don’t underestimate the work in creating the content + customer expectations @LizzieMaguire @ANZ_NZ @BusinessCircle_ #thenetworkeffect @Soulmachines #AI pic.twitter.com/v5uuC52KwX— Divina Paredes (@divinap) April 2, 2019
Get the latest on digital transformation: Sign up for CIO newsletters for regular updates on CIO news, career tips, views and events. Follow CIO New Zealand on Twitter:@cio_nz