Australia Post is six weeks into its first GPT experiment

Australia Post is six weeks into a proof-of-concept of a GPT-enabled virtual assistant that is already judged to be “really promising” as a frontline customer support tool.

Gary Starr talks generative AI.

Executive general manager of parcel, post and ecommerce services Gary Starr told the Online Retailer Conference in Sydney that generative AI appeared transformative, and so “sitting [it] out is not an option”.

Starr said that generative AI appeared to have particularly strong use cases for frontline customer service, but added that “like the best of human-powered frontline interactions, the value has to be delivered with empathy and with humanity.”

“Australia Post is testing this technology first in its contact centre, that’s where you’ll see a lot of prominent work across the world in generative AI,” he said.

“We kicked off a proof-of-concept about six weeks ago to evaluate how we can use AI to address customer queries. 

“We’re moving from a traditional chatbot to one that doesnt require the structured data to feed it, that thrives on generating new media and sounds as human, or more human, than humans.”

That appeared to be a major selling point for the technology for Australia Post - that it offered a shorter time-to-value than traditional chatbots, which require a lot of intensive training, upfront and ongoing, to be able to respond accurately to whatever a customer might ask, and all the different ways they might phrase a question.

“We didn’t have to feed this [with] specific use cases, it really was up and running in a matter of weeks,” Starr said.

“Early results have been really promising and we’ll look to continue to develop this.

“It’s still early days and the technology continues to evolve, and we’ll have to evolve with it.”

Starr showed a pre-recorded demonstration of the GPT-powered virtual assistant in action, where a customer was seeking information about a parcel delivery. 

He particularly highlighted its apparent natural-language abilities, which included being able to switch languages in a single chat - if a customer suddenly started to speak another language.

It isn’t clear how the GPT-enabled virtual assistant is currently engaged, nor on what technology foundation it has been developed.

The hope, though, is that it can further ease the workload on human agents by handling more query types, leaving agents to handle only the more challenging support requests.

“What we’re seeing with generative AI is an opportunity to replicate the best of human frontline experience with digital,” Starr said.

“This means using data and analytics to have an ongoing conversation with customers. 

“It enables us to understand where and how we can add value at each moment. This could [also] be an opportunity for engagement for proactive servicing or providing a customised offer.”