Ash—the AustralianSuper website helpbot—has been a hit with members and staff alike since her January introduction. So much so, that she is being promoted; no longer just a chatbot, Ash will soon also be a ‘voice bot’.

On the fund’s website, Ash initiates conversations by displaying the words ‘Hi, what can I help you with today?’ in a chat window. With voice bot Ash, it’s the member who’ll speak first—literally. Anthony Gurciullo, AustralianSuper Digital Services Manager, explains:

“To talk with Ash, you’ll say ‘Hey Google, connect me to AustralianSuper’ within earshot of a Google Home device. Ash’s Australian accented reply, ‘ask me a question’, will enable the conversation to flow.’

Immediate impact

Ash is built with Google Dialogflow, a natural language interface. “Dialogflow uses artificial intelligence to interpret, and answer questions”, Gurciullo says, “the technology can be applied to both spoken and written conversations”.

Upon introduction to the AustralianSuper website, chat bot Ash immediately began answering 60% of frequently asked member questions on her own. Today, that number is nudging 75%. “We’ll never hit 100%, but 85% is definitely achievable”, Gurciullo says, adding “we’re still learning, Ash is learning, but even at 75% the benefits are enormous.”

Website questions Ash can’t answer are transferred to a contact centre operator, and the chat window conversation continues. To save time and repetition, the operator automatically receives details of the preceding exchange.

A fast learner

When Gurciullo says ‘Ash is learning’, it’s no mere figure of speech. At the completion of their exchange with chat bot Ash, users are invited to complete a short survey. Employing complex algorithms, Ash uses the results along with other data to self-assess and improve her knowledge base automatically.

On the odd occasion Ash handles a question poorly, the conversation transcript is reviewed by Gurciullo’s team, who then update the knowledge base manually. “That aspect is time-consuming”, he admits, “but is far outweighed by the better utilisation of contact centre resources”.

In her first nine months of deployment, chat bot Ash handled roughly 21,000 conversations successfully and handed over another 7,000 to contact centre operators. Factoring in the time spent on preliminaries and small talk, Gurciullo estimates that provided operators with an additional 1,000 hours in which to help members with complex issues.

“Our contact centre has adopted Ash like they would a human team member”, says Gurciullo, “if we switched her off now there would be an uproar, because Ash frees operators from the more mundane contact centre interactions”.

Lost in translation

Although a native English speaker, Ash occasionally struggles with the Australian vernacular, and also with the tendency of some members to pepper written exchanges with SMS abbreviations. And of course, not everyone can spell. Ash always has a plan B, however.

“If she doesn’t understand a question the first time, Ash will ask the member to pose the same question in a different way”, Gurciullo explained. “If she fails a second time Ash will propose handing over to an operator or, if the exchange takes place outside normal hours, that an operator contact the member on the next business day”.

Personality was another challenge. “Early on, we programmed chat bot Ash to show a little ‘spark’”, Gurciullo noted, “but she lacked appreciation for nuance, and reacted with inappropriate zeal to questions where sensitivity would be expected, such as financial hardship”. The remedial learnings were swift, however, and Ash quickly emerged the wiser.

The future in context

The next goal for Ash is ‘contextual holding’. At the moment, she interprets each question in isolation. Contextual holding will enable Ash to recognise when the second question asked relates to the first, enabling people to have more of a natural conversation.

Further innovations are also in the pipeline. “The fund receives 300,000–400,000 email enquiries annually, each of which must be read and responded to individually”, says Gurciullo.

“This time next year, we aim to be applying artificial intelligence to handle the bulk of them more efficiently. We’re also looking to introduce a version of Ash specifically for employers.”