At this stage, Agent Pay is what Mastercard calls “human in the loop” technology with “Book for me” and “I authorise this AI agent to make this purchase for me” buttons that need to be clicked.
But Mastercard is also planning a fully autonomous option, where an AI agent can buy things on your behalf.
It will offer the ability to pre-set a price ceiling if you say, want an agent to join a queue for concert or sports event tickets but are wary of a Fifa World Cup-style dynamic pricing blowout – although even then there can be many nuances, with so many different seating options, only some which a human would pay over-the-odds for in the heat of the moment.
A key reason to use your credit card for online purchases is the full refund guarantee – in most cases – if something goes wrong.
“If you buy something and it doesn’t come, or if you buy something and it turns out to be a scam website, you have zero liability as a consumer,” Simons says.
That promise also includes “human” in the loop Agent Pay transactions.
What about a transaction made by a fully autonomous agent? Will those sorts of transactions have the same zero liability?
“Yes they will,” says Simons.
“That’s our brand promise. The important thing is going to be trust. If you’re paying on a Mastercard, it doesn’t matter if it’s an agent or a ‘future intent’ agent or online or in a store, you have zero liability and that will absolutely stay in place.”
Last year, when the Herald polled various players, policies about AI agent purchases were fuzzy around the edges. Simons has crystalised her company’s position.
She adds that once autonomous agent buying is here, you’ll also be able to review a list of your “future intent” purchases, be they a flight to Wellington as long as it’s below $300 and before 9am, or a pair of black denim pants from a particular retailer if the price falls below $150. They can be tweaked or deleted.
When will it go mainstream?
Simons expects “human in the loop” Agent Pay to go mainstream over the next nine months or so as more partners come onboard at every level.
Mastercard is in talks with the other major banks (similar showcase transactions in Australia earlier this month also included Westpac, plus ASB parent CBA).
On the LLM (large language model or AI chatbot) front, while both the Aussie and NZ transactions featured an agent made by Maincode (the company behind Australia’s sovereign AI, Matilda), Simons says “globally we are working with all major players – including OpenAI, Microsoft, Google”.
Multiple merchants are also being lined up.
Recognising agents
While she talked up scam protection, Simons also said her firm’s Agent Pay framework, which builds on its existing security token effort and Click to Pay technologies, is initiated with a “handshake” between the retailer and the agent, ensuring everything’s above board.
As well as preventing fraud, the “handshake” process is also important for reducing false-positives – or AI transactions being declined because an agent seems dodgy.
Simons says an agent is very rapid – a characteristic shared by the bots deployed by scamsters.
“It doesn’t look like a normal human behaving in a normal way.”
Agent Pay helps to establish an AI agent is registered and legitimate, and that it is following a customer’s authorised intent (which also helps protect a retailer from claims of unauthorised use by a card holder).
She says retailers and service providers of all stripes have to prepare for a world where many – and ultimately most – visitors to their sites and apps are AI agents, who behave differently, and customers who, like herself, are now more likely to hit up an LLM like ChatGPT with a shopping query than hit Google.
Mastercard can work with big retailers on their “discoverability”, Simons says.
She adds, “for small businesses as well, it’s not going to be about how many dollars you’re spending on Google AdWords anymore. It’s about, how do you become more discoverable for these agents”?
‘Addresses a genuine gap’
“People are using AI tools to search, compare and shortlist purchases right now. Most of the commerce ecosystem cannot see when that is happening. From a merchant’s system, an AI tool browsing a catalogue looks identical to a bot,” says Deloitte NZ AI director Dr Amanda Williamson.
“From a bank’s perspective, it looks like unusual activity. The concept of making AI-initiated transactions identifiable across the payment chain addresses a genuine gap.”
Williamson sees another pressure for retailers: pricing. A traditional search engine makes it easier for a person to find the best price. An AI agent will get endless quotes.
“Business models built on human effort constraints will face pressure,” she says.
“Businesses that are only legible to humans will become harder to find. When software does the browsing and purchasing, it interacts with structured data, pricing logic and policies. It does not read a web page the way a person does. If your catalogue is not readable by an AI tool, your products do not get surfaced.”
And although Mastercard customers will have zero liability for transactions under its Agent Pay framework, Williamson says that more broadly,months it’s an open question whether our consumer and commerce law is ready for the new era AI agent buying.
Room to finesse
The rise of agentic AI has seen various companies release agents that can buy things on your behalf over the past few months.
“I have been experimenting with agentic AI shopping tools since they became available. They can be extremely erratic,” Williamson says.
The agents that require human confirmation are too slow and clumsy to be worthwhile.
And fully autonomous agents have been outright perilous.
“I asked an AI to find me a T-shirt. It found 50 identical ones, added them all to my cart, and tried to process the payment. Had I not caught it in time, I would be explaining to my family why I spent over $800 on ‘I [heart] AI’ shirts,” Williamson says.
Short, snappy, maybe more choices needed
Simon’s Agent to Pay transactions were short and snappy.
Mastercard has built on existing standards and technologies, including tokenisation (replacing a 16-digit number with cryptographically secure credentials ensuring each transaction is traceable and authenticated) Apple Wallet and Google Wallet, PassKey and Click to Pay.
They were arguably too short and snappy.
Not many will be comfortable picking upmarket Queenstown accommodation based on a choice of one of three thumbnails.
And any penny-pinching Event Cinema goers, like your author, would like to log on to the theatre operator’s Cinebuzz loyalty scheme to get a couple of dollars off the price of the tickets and a cheaper coke and popcorn.
“Building loyalty into those experiences will be really important. It will happen,” Simons says.
Filling in such features will ultimately happen on the app maker and retailer’s side, but Simons says her firm’s Agent to Pay framework makes things more transparent.
Chris Keall is an Auckland-based member of the Herald’s business team. He joined the Herald in 2018 and is the technology editor and a senior business writer.
