EXCLUSIVE: "Agent of Change" - Michele Centemero and Eimear Creaven, Mastercard in 'The Paytech Magazine'

Mastercard is responding to the consumer-driven revolution in agentic commerce by building a new trust architecture

As consumers, we’re used to having our identity checked at the point of sale (POS). An entire regulatory system and the technology that supports it have been built around POS and know-your-customer (KYC), be it online, on the phone or in person.

But, in future, merchants and their payment processors won’t be asking who is behind the payment, so much as what. And, by extension, did it have the authority to make a transaction?

“We are entering an incredible moment in payments when we move from point-of-sale to point-of-intent,” says Eimear Creaven, President of Global Partnerships at Mastercard. 

Mastercard revealed earlier this year that it had been working with Google on creating a Verifiable Intent Architecture, including a cryptographic audit trail linking user identity, AI agent instructions, and transaction outcomes in preparation for a new era in non-human and agent-to-agent payments.

Consumers leading the trend

Almost one in two (45 per cent) of consumers already use AI for at least part of their shopping experience, according to IBM research from January 2026. But with the rapid pace of change, today’s figure is likely to be much higher… and higher still tomorrow. Amazon’s July 31 2026 earnings call revealed that 350 million customers were using Alexa For Shopping’s personalised recommendations, product comparisons, price history and automated shopping= features, like price alerts and auto-buy.

Engagement had almost doubled during the previous quarter alone. That figure wouldn’t surprise Creaven.

She says consumers feel more comfortable using agents with transactions that are repeatable, ‘where they can really build confidence’. This makes restocks or auto-buys the perfect vehicle for trust-building, priming customers to hand over more of the reins of their purchasing decisions.

The pace of mindset change is truly fascinating, and perhaps unprecedented in the fintech world. Ingrid Lommer, Co-Founder of Marketplace Universe, uncovered that the number of people in Germany willing to let AI agents take over their shopping jumped from 10 per cent in December 2025 to 35 per cent in March. More than a three-fold leap in as many months. Another June 2026 study by Accenture revealed that 74 per cent of us would trust an AI agent to buy for us over our best friend.

Convenience wins out

“Consumers are always looking for something that simplifies their journey,” says Mastercard’s Executive Vice President, Services, Europe, Michele Centemero. “AI and the agentic era really free up time for people. Usage and adoption will be fast and huge.”

We’ve seen consumers lean towards simplification time and time again. From the rise of the supermarket over family-run grocers or butchers, to the success of e-marketplaces over shopping malls. Why should e-commerce be any different? Bain & Company estimates that AI agents will be behind 15-25 per cent of all US e-commerce sales by 2030, creating a market opportunity worth $300-$500billion.

But while consumer demand rockets, evidence shows merchants are not keeping pace: 45 per cent of customers use AI to help them shop, yet Salesforce found just 28 per cent of merchants had implemented AI into their processes. This creates a huge gap, funnelling customers to better-adapted competitors. To thrive in a generative AI shopping world, merchants must be more than simply online. They need to be plugged in to agentic commerce infrastructures, data-rich and with catalogues that can speak the language of AI shopping agents.

Creaven believes businesses that have pre-existing AI capabilities will find it ‘a very easy move from AI-assisted to fully autonomous’. In turn, this is likely to lead to ‘an opportunity to really run at scale’.

Investments over the past years in AI tools, hyper-personalisation and data security should pay for themselves in gains of visibility, agility and efficiency. But merchants that don’t make the investment will feel the hit. We’ve seen how sudden the fallout can be for those slow to adapt to new shopping trends, methods or channels. The early noughties are littered with examples of seemingly indestructible companies that failed to adapt for e-commerce.

Borders bookstore paid the price for its lacklustre reaction to the internet. It’s the same story for the likes of Toys R Us and Blockbuster. Agentic commerce looks like the next retailing revolution and SMEs need support. That’s what Mastercard is mustering. Take Shopify, one of the biggest and most dominant e-commerce platforms in the world, powering more than 4.6 million active websites and handling hundreds of billions of dollars in yearly sales. It is one of Mastercard’s many integrated platforms, allowing merchants to benefit from its agentic payment capabilities.

Most (95 to 98 per cent) of Shopify’s customers are small to medium businesses, with fewer than 10 employees. The e-commerce platform is committed to future-proofing its B2B clients for agentic commerce, and one of its initiatives is the creation of a merchant data protocol, designed to help AI agents read product and service catalogues accurately.

While merchants will still have plenty of work to do to stay relevant, projects like the merchant data protocol give them a sporting chance to be seen and compete in the agentic commerce world.

 As Creaven explains: “We have to ensure that all consumers and all merchants, big and small, are included in this journey.”

“We need to grant visibility to small businesses in this arena,” agrees Centemero, ‘to avoid four marketplaces taking over everything.”

When retailers invest in agentic shopping technology, it’s a very smart move. May 2026 research, also from Bain & Company, found US shoppers are 3.5 times more likely to trust a retail-owned AI tool than an AI platform like ChatGPT (25 per cent v 7 per cent) to make a purchase.

Opportunity knocks for banks

Consumers are even more likely to trust their incumbent bank, which provides a unique window for institutions to reclaim relevance. In a showstopping live demo at Money20/20 Europe in June, Mastercard, Worldline and ING demonstrated how that was possible, using Mastercard’s Agent Pay to make an end-to-end agentic transaction, direct from a banking app. 

Within the ING app, a consumer instructed the AI agent to find tickets for a romantic concert in Amsterdam. The agent quickly came back with some options and prices. From there, the customer gave permission for the agent to purchase two concert tickets.

It was a glimpse of the future and, as account-to-account payments increase, it’s an outright opportunity for banks to support the agentic shopping trend. 

But what happens when things don’t go as smoothly? Who is responsible for issuing refunds or chargebacks in a scenario where AI agents may be interacting with other AI agents? And how do we know an agent didn’t go on a rogue shopping spree?

Those are the questions around trust and accountability that Mastercard’s Verifiable Intent infrastructure, which underpins its agentic payments systems, including Agent Pay, is designed to address.

Using tokenisation and blockchain principles, Verifiable Intent creates an immutable record of what interactions and authorisations took place, when and by whom. The infrastructure establishes a shared source of truth across the whole ecosystem, tracking cryptographic proof of authorisation, which could be relied on by consumers, merchants and issuers in the event of a dispute.

“It’s not only a matter of knowing your agent,” says Centemero. “It’s more the moment in which you know the intent.” Mastercard’s Verifiable Intent gets us to= establishing dependable permissions and accountability for agentic commerce. But it cannot solve every issue, because intent is inherently subjective and humans are irrational.

The consumer who instructed their agent to ‘Buy me two Sting concert tickets in London’, might have forgotten to add, ‘anytime except August, because I am on holiday then’. Or ‘only if the seats are together because I don’t want to be sitting on the other side of the stadium from my teenage granddaughter’. Or ‘Did I say Sting? I meant the frontman from Dire Straits, I always confuse them!’.

This is where hyper-personalisation and rich data elegantly collide. Blended with technology like Verifiable Intent, merchant data can plug much-needed gaps. It means AI agents will need to know their human better than the relatives who realise that when grandpa says ‘Sting’ he means ‘Mark Knopfler’.

Compliant access to calendars, previous purchases, tastes and preferences can help to reduce mistakes and accountability issues, by at least throwing up a query. That said, all ecosystem players should acknowledge that not all intents will be adequately ascertainable, and price this into their strategy!

Interoperability is urgently needed

Verifiable Intent helps to build confidence in agentic commerce.

“Trust is a critical factor to help us scale agentic,” says Creaven. “But married to that, to ensure that the user experience is smooth, we need interoperability.”

This allows consumers to stay on the platforms they love and trust, while also benefitting from the same seamless service that will hopefully become standard across the industry, integrating through open APIs. There are already several open standards in development, notably the Agentic Commerce Protocol and the Universal Commerce Protocol, launched by Google. These allow any business or AI platform to implement a code so they can participate in some degree of agentic commerce – for example, allowing buyers to find the merchant’s products with an AI search. 

Some payment service providers have already integrated these codes, including Stripe, which has partnered with the Agentic Commerce Protocol, automatically giving its business customers more visibility. As we move into the next era of AI agents speaking to other AI agents, sharing a common language will become business-critical.

Potential for increased cyber risk

Centemero also wants to see the industry share more knowledge in order to stifle fraud. Palo Alto Networks has anticipated some ways that criminals may commit fraud using agentic commerce. In one scenario, it imagines a shopping agent visiting the criminal’s site to check for discounts, and an indirect prompt injection reprogrammes its memory. This could instruct the agent to send funds to the criminal.

Another scenario looked at criminals training their own agents to perform requests using hidden HTML metadata or spam tags. The rise of internet shopping, especially that assisted by AI, has also already seen a spike in returns fraud. Clearly, agentic commerce could open new avenues for bad actors.

“We need to all work together to share information, knowledge and capability in order to be two steps ahead of the bad guys,” says Centemero. “Our focus in security is to assess the transaction, to protect consumers and merchants, and to prepare the market. Preparing the market means also talking to government, corporates and consumers because one of the advantages we have as an industry is that we can talk to each other – the fraudsters can’t.”

Agents and tokenisation

What Creaven describes as ‘a huge shift in how consumers think about entering the payment cycle’ has come at a time of another significant – but synergistic – disruption.

“We’re also seeing the global expansion of tokenisation, particularly of digital currencies and stablecoins,” she says. “When you think about how consumers or small businesses are going to engage in commerce at the front end and how those transactions are going to settle at the back end, the opportunities of bringing these trends together are going to be huge.