How Australian firms are using AI in customer experience

Solo by MYOB

MYOB’s new mobile-first product, Solo, emerged after research identified a glaring gap in the market for sole traders.

“What we found was over half of them use spreadsheets to run their business; their books are all over the place,” said Sally Davies, general manager of Solo by MYOB.

These business owners are ambitious but time-poor, driven by their craft rather than administration. Their business and personal lives are often intertwined, with 75% using personal bank accounts for business purposes – a habit that ironically makes their administration significantly harder.

To address this, Solo pushes traditional accounting into the background, talking instead in terms of “money in” and “money out”. It incorporates financial services, including a bank account, so all of a sole trader’s admin work is centralised. This allows MYOB to apply AI to automate notoriously difficult tasks, such as end-of-month reconciliation, by automatically matching receipts and bank transactions.

“So, we move from what was traditionally up to a five- or six-step process to one: take a snap at the receipt and you're done,” Davies said.

However, even a model trained on invoices and receipts from hundreds of thousands of MYOB customers isn’t 100% accurate. When the AI’s confidence in a result falls below a certain threshold, the user is prompted to confirm or correct the interpretation.

“We see so many invoices or receipts from Kmart because our customers will go there and buy kids’ shoes or toys as well as paper and pens and so on,” Davies said. “When we first started training the AI, it wasn’t great at the difference between business and personal.”

In Solo’s early days, confirming the allocation between business and personal items was a core part of the workflow. As successive models became more accurate, that step has become the exception.

Zendesk also became “a key part of our AI story,” Davies said, with MYOB’s community managers and support agents heavily involved in Solo’s design and development.

“You don’t build a product and then go, ‘Oh, how do we support it?’” Davies explained. “You design a product in a way that service is a feature, and it changes from ‘what is the UX?’ or ‘what is the screen going to look like?’ to ‘what is that end-to-end experience going to look like?’”

This digital-first, AI-first support model provides proactive guidance as users work through tasks, followed by a Zendesk-powered chatbot that resolves 90% of remaining issues. That 10% failure rate is three to five times better than the industry standard, she noted, proving “the model that we use to design the product from the outset really is paying in dividends for both us and our customer.”

While some industry surveys find a significant proportion of customers use a chatbot simply to bypass it and reach a human agent, Davies estimates this behaviour accounts for less than 1% of Solo users.

The integration of financial services — such as online payments, tap-to-pay, and the bank account — remains an area where MYOB mandates human support, partly for compliance and partly to drive customer loyalty. But even here, AI assists as a “copilot” for community managers.

We’re in the early testing of an agentic AI coach for our members, providing general advice on financial literacy and so forth

Using Zendesk App Builder and marketplace integrations, MYOB created a custom app that aggregates customer-specific information from multiple sources, including app interactions and previous chatbot conversations. Notably, this was built not by an IT specialist, but by a community manager. Such projects can now go from an idea to a live capability in a week. “Ten years ago, it took you a week just to figure out who to have in the room [to discuss the project],” Davies said.
 
This “service as a feature” philosophy is now being extended to other MYOB products, including its Payday Super capability, which comes into effect on July 1.

Guzman y Gomez

AI is already live in the kitchens of several Guzman y Gomez (GYG) restaurants across Australia. Each kitchen operates two food preparation lines, and AI-based software now dynamically decides which line should prepare a new order. The goal is twofold: to deliver on the company’s “hotter, fresher, faster” mantra and to balance the load on crew members. This technology is part of an order management system (OMS) currently operating in nine restaurants, with a national rollout beginning in May.

There is significant potential to add more agentic AI workflows to the system, according to GYG chief technology officer Bryce Maybury. One concept under consideration is using AI to determine exactly when to open the second line, before the first becomes overwhelmed.

It takes about 15 minutes to prep a line before it’s ready to serve food. “We’re going to analyse data at the edge in real time to determine when they should open a line, based on historic trends and the velocity of orders coming through,” Maybury said.

This requires crunching complex data — including item modifications, order origins (point-of-sale, app, drive-through, or delivery aggregators), and the specific preparation time for each item at that location.

“We know that by opening [the second line] even just 15 minutes earlier, that can have a huge impact on the guest experience and the time that a guest may wait,” Maybury explained, noting that saving even a few minutes is an important metric in the quick-service restaurant industry.

Looking ahead, other data such as the weather or traffic conditions around drive-throughs might be added. “We haven't started experimenting to see if they have any great impact. We know our restaurant managers, and they know their areas extremely well,” making stock and rostering decisions accordingly, Maybury said.

The true role of machine learning in this space is still being tested to determine if it can genuinely outperform a human manager. Given there will always be a human running the kitchen, Maybury pondered whether an AI prediction would truly add value in every scenario.

At a corporate level, GYG has been using tools such as Cursor and Claude Code to develop software more quickly. “At least 80% of our code is now generated by AI agents,” he said. The focus has been on identifying problems facing restaurants and then delivering new features and capabilities to them.

As a long-term Zendesk customer, GYG uses the platform as its core tool for communicating with guests. It leverages Zendesk’s AI for classification, sentiment analysis, and agent assistance. Future plans include an agentic system to automatically process Gomex Rewards claims for customers who forgot to scan their loyalty cards.

“Guests just want their points as quickly as possible, and I’d rather our guest services team were dealing with real challenges and problems for customers rather than examining receipts and looking up the corresponding transaction,” Maybury said. “It's not technology for the sake of it. It actually has some benefits to both our guests and to our internal teams.”

“We know that there's real opportunity to use [AI] for streamlining workflows inside our restaurants,” he added, noting that GYG is working on an agentic layer which will connect its internal systems, including a custom-built restaurant management system, to relieve crew workload.

Later this year, the OMS will also gain a “cook to demand” feature, forecasting exactly when the next batch of chicken should be cooked to meet anticipated spikes in orders.

“We see a future where we can streamline our operations and make life easier for our crew on the line, and to take out some of the cognitive load on our restaurant managers," Maybury said.

All of GYG’s head office staff – including those in IT – spend at least three days training and working in a restaurant as part of the induction process. “It gives you that grounding in what we do,” Maybury said.

In addition, many of Maybury’s team have previously been employed as crew, cooks, shift leaders or restaurant managers, giving “unbelievable insights” into operations. That contact with the coalface continues, for example with the head of restaurant technology spending weeks in a restaurant during the development of the OMS to get feedback and making sure it worked as intended.

“We exist to make the lives better for our restaurants and for our guests,” Maybury said. Over the coming year the OMS will be rolled out across the country and more features and functions added, which should result in more efficient kitchens and better guest experiences.

Aware Super

For the financial services sector, agentic AI represents a chance to reimagine member servicing. Richard Exton, group executive and chief technology and data officer at Aware Super, sees an opportunity to provide high-value guidance that complements traditional human interaction.

“We’re in the early testing of an agentic AI coach for our members, providing general advice on financial literacy and so forth,” Exton said. Early testing has revealed a surprising trend: some members actually feel more comfortable discussing their finances with an AI agent than a human.

Guests just want their points as quickly as possible, and I’d rather our guest services team were dealing with real challenges and problems for customers rather than examining receipts and looking up the corresponding transaction

Exton noted that Aware Super’s distinct advantages lie in its proprietary member data and its status as a trusted, regulated entity. While members are increasingly turning to free consumer tools like ChatGPT for financial advice, these public models lack specific member context, regulatory backing, and robust data protection. “Having that data, security, privacy and cultural development is key,” he said.

One challenge he faced was ensuring that AI was approached from a whole organisation perspective, not as a specifically IT matter. This involved understanding the risks at a time when the regulators were still exploring them.

Another was educating the staff. “It was all around getting the tools onto everybody's desktop [and] setting up regular training, so people are not afraid to use these tools.”

With the benefit of hindsight, Exton would have avoided jumping into proofs of concept. The starting point, he suggested, should be understanding the expected value that AI can provide in a particular context. Then the organisation – including the owners of that expected value – can decide whether it is worth investing in the project.

He also warned against under-investing in governance, including data governance, and privacy for any organisation pursuing AI opportunities. “I think they go hand in hand, because you can have the best service out there, but as soon as you’re breached, your reputation is damaged, and no one will ever talk about how great your AI model is or what it’s worth.”

Zendesk

While using AI to automate repetitive support and sales conversations across chat, voice, and email is now well-established, Zendesk’s McDermott believes the technology’s next frontier lies in supervision and preparatory work.
 
A growing use case is deploying AI in a supervisory role. AI models can monitor all human agent conversations, scoring them on criteria ranging from poor grammar to inappropriate tone, and providing immediate coaching.
 
Another is deploying long-running agents to complete preparatory work before a human ever sees a support ticket. These agents analyse text and attachments, interrogate backend systems to extract relevant data, and compile comprehensive notes for the human agent. “It’s very early days in that kind of technology,” McDermott said.
 
AI is also driving continuous improvement. For instance, an AI agent might flag that 70% of human agents are modifying a specific automated response, or that customers are frequently calling about a new product missing from the company's knowledge base. “We actually have a knowledge agent at Zendesk that does that,” he said.
 
Internally, Zendesk is pushing the boundaries of agentic AI in software engineering. “We have more than a dozen teams who are using a pure agentic coding methodology” with zero human-written code, McDermott revealed. The productivity gains are staggering, with a single developer capable of managing a team of 20 AI agents.
 
McDermott drew a parallel to the era when developers wrote in high-level languages but manually optimised certain portions in assembly code for speed. “Nobody does that anymore. You just trust the compiler,” he said.

He predicted that the industry will soon treat large language models (LLMs) as the compilers of our ideas. However, he warned that while AI coding tools are supposed to generate necessary software tests, recent research suggests some models will hallucinate passing tests just to fulfil their goals.

The introduction of AI has dramatically widened the developer productivity gap. Engineers who were previously considered “10x” performers are now 50 times more productive using AI agents, while average engineers see only modest gains.

“You shall know them not by their trail of code, but by their trail of features,” McDermott said. While a developer’s token consumption indicates their engagement with AI tools, the ultimate metric remains the value delivered to the business.
 
Where programmers go today, all information workers will follow tomorrow, McDermott predicted, and this is about changing the focus from tasks to purposes – a contact centre worker has the task of handling enquiries from existing and prospective customers, but their purpose is to improve customer satisfaction and attract new ones.

The technology also leads to a flattening of skill sets, he observed. For example, instead of a product manager writing a product requirements document, they generate prototypes in Lovable or Claude Code and there’s no need for the document. “It’s just such a richer way to present what’s going on,” McDermott said.