Mastercard has completed a series of agent-led commercial transactions in Australia and New Zealand to test how artificial intelligence (AI) can move beyond recommending actions to executing them.
While previous experiments this year focused on consumer purchases, this latest round focuses on commercial payments and everyday decisions handled by small and medium-sized businesses. The transaction was completed through a combination of banks, merchants and platforms. These include Hnry, MYOB, Pay.com.au and more.
Related Articles Block Placeholder
Article ID: 334821
Mastercard says the goal is to connect the currently separate steps of inputting business data into an AI system, generating recommendations, and completing payments. The difference is that these deals combine insight and execution, rather than the business owner himself acting on the recommendations.
“There’s still a disconnect…you have to leave the AI platform and go to that merchant to make a payment,” said Anuska Lads, executive vice president of commercial and new payment flows for Asia Pacific at Mastercard. smart company. “That last mile is where agents play a role.”
What this means for small businesses
For small businesses, relevance comes down to time and how you get things done.
Mastercard says small businesses spend about 60 hours a week running their business, and an additional 10 to 20 hours on administrative tasks such as payments and financial management.
“Small businesses…don’t have that passion to be a CFO, right? In fact, being able to take that away and help them make that happen…gives time back to small businesses,” Lads said. Smart company.
Mastercard’s demonstration emphasized its focus on everyday financial managers. In one example, business owners looking at upcoming bills were prompted to consider when and how to pay based on cash flow and potential compensation. In another case, an agent identified low inventory levels and triggered a purchase from an existing supplier.
Related Articles Block Placeholder
Article ID: 330566
In another demonstration, an agent presented an upcoming bill and calculated how to maximize points by transferring the payment to the card before completing the transaction.
Mastercard Australia senior vice president Surin Fernando said the aim was to reduce the number of steps required from identifying a task to completing it.
“What we’re most excited about is when we can combine that data with AI to drive actions that may lead to payments,” Fernando said. Smart company.
Constraints behind the Mastercard pilot
This framework makes sense, especially for companies that are already juggling multiple systems. However, the transaction shown here was completed under controlled conditions, with banks, merchants, and software platforms working together.
“This is an ecosystem play…no player can do this alone,” Lads said.
Relying on multiple stakeholders will determine how quickly this can move beyond testing. Banks need to support authentication and payment flows, software platforms need to integrate AI into existing products, and merchants need to have access to inventory and pricing data in a structured way so agents can act on it.
Related Articles Block Placeholder
Article ID: 314553
The system also relies on underlying AI models and platforms that Mastercard does not control, adding another layer of dependence for businesses.
Fernando said that there are still necessary measures across the value chain, including upgrades on the acquirer side and further preparation on the part of merchants.
It also raises the question of how these systems are delivered to businesses. Rather than existing within a single product, the agent experience could appear within accounting software, banking apps, or other tools small businesses already use, depending on the implementation method chosen by the partner.
It’s also unclear what the costs of these features will be as small businesses move beyond the pilot stage.
Still, Mastercard positions the technology as relatively close.
“That’s definitely something in the near future.” [opportunity]… It’s not something that’s going to take two or three years,” Lads said.
But for now, the technology is still in the testing phase, and broader deployment will depend on how quickly these pieces are integrated. It also raises practical questions about liability and error, especially if the system makes technically correct but commercially incorrect decisions.
