Transurban explores bringing agentic AI to its chatbot

Transurban is exploring how to build on a year-old generative AI customer service chatbot, giving it contextual information about customers, with a longer-term goal to have the chatbot proactively engage with customers directly.

The toll road operator introduced a generative AI chatbot, built on Anthropic’s Claude large language model, in August last year on is Linkt app and website.

Linkt is where customers sign up for a toll road tag and manage their payments. It’s also a key digital channel for customer support.

General manager for customer experience and operations Chris Jackson told the AWS Summit Sydney that Lex is currently able to respond to customer questions 24×7 by drawing on the content in the “help” part of the site and app.

This has seen it assist people with e-tag issues or with information regarding natural disasters such as Cyclone Alfred.

The chatbot was built by services firm Concentrix Catalyst and AWS. Amazon Bedrock is used in part to implement guardrails around how the chatbot can be used, including what topics it can handle and what it needs to break out to a human agent immediately.

“The Bedrock platform provides a really robust environment to enable you to put clear guardrails in place around the sorts of questions that you either just don’t want the chatbot to engage with and … the sorts of questions that you don’t want the chatbot to try and solve because you actually want to refer them to a human agent,” Jackson said.

“For example, customers experiencing financial hardship or family and domestic violence, who are experiencing challenges with their tolls, we want to be able to refer them immediately to a customer service agent as opposed to the chatbot handling those queries.”

Jackson said that in addition to answering customers’ questions 24×7, Lex had proven capable in getting customers connected to human agents much faster than when customers had to navigate a complex menu of support options.

“If you ask a question that Lex knows needs to be referred to an agent, you’re immediately referred, and equally, if you ask to speak to a customer service agent, you’re immediately referred,” he said.

“We are getting customers to humans who need to support them much faster than we ever did with the sort of traditional menu based model where customers had to navigate through that excruciating process, even if they knew they really needed to speak to a person.

“That ability to really help customers, but also get humans in front of people when they need it most, was really important to us in terms of thinking about how we build emotion into the broader customer experience.”

The toll road operator is now planning several future iterations of Lex.

The most immediate desire is to give Lex some contextual information about the customer asking a question, so it can better tailor a response or suggestion to fix a customer’s specific issue.

“In the next iteration of the chatbot, what we’ll actually do is provide Lex with access to contextual-based information about the customer: what their account balance is, what state they’re in, is their account active or suspended, what product they have,” Jackson said.

This will lay the foundations for Lex to do even more, via agentic AI-based services.

Eventually, the goal is for the agentic AI to be capable of offering proactive customer service – to recognise a potential issue with a customer’s account and get in touch with the customer with a resolution.

This, Jackson said, “requires more investment in guardrails, more investment in testing and a greater understanding of the various different scenarios and the ways in which the agent could interact with a customer.”

But he noted that “the value to customers becomes exponentially [more] powerful.”

“There’s the opportunity for Lex to reach out proactively to me and say,‘It looks like you might be on the wrong product based on an analysis of your trip history, we think you’d be better off with this product’.”

Lex could also offer to update a customer’s account if the roads they are travelling suggest they have moved interstate, or to check whether they are eligible for a cashback offer based on their location.

“You can see the opportunities in that context are endless,” Jackson said.

Transurban is still thinking through some of the challenges with a proactive AI-based engagement model, however.

“At the moment, we are very declarative with customers that they are engaging with an AI-powered chatbot,” Jackson said.

“We see overwhelmingly that customers have a great experience with it, but you can see as you flip that to proactively engaging customers it just lifts the level of complexity and the level of challenge that little bit further.

“The customer hasn’t chosen to interact with the organisation in an AI context [in this way].”