From top of wallet to top of model: how AI will reshape payments

Mariagiovanna Di Feo, Partner, Bain & Company (EMEA Payments): As AI agents begin to influence what consumers buy and how they pay, the payments industry faces a fundamental shift in trust, competition, and who controls the decision. 

AI is increasingly being embedded across the payments value chain, from fraud prevention to customer interactions. Where do you see AI delivering the most significant impact on payments over the next few years?

I would separate two things: the first is the AI transformation of the payments company itself. That is where most of the industry conversation sits today, and it is already happening: fraud, disputes, customer service, merchant onboarding, Know Your Customer (KYC), reconciliation, software development, sales… Payments is data-rich and operationally complex, so the productivity and service upside is real. But, at some point, everyone will get those gains. They become table stakes, not advantage.

However, to me, the more interesting disruption is the second: AI changes the actual commerce and payment decision. Today, customers decide what to buy, where, and which card or wallet to use. An agent will be able to increasingly do all those things for the customers: search, compare, recommend, build the basket, select the method and eventually execute within the rules they have set. That is a profound shift, because payments have always been designed to influence a human being during checkout.

For decades, the goal was to be top of wallet — make your card attractive, habitual and visible enough that the consumer chooses it almost automatically.

Now, in an agentic world, the goal is to be top of model. The agent can compare options on every single transaction — rewards, merchant fees, available credit, protection, convenience, even the likelihood of a dispute — and optimise accordingly.

As AI becomes more capable of making decisions and initiating actions on behalf of consumers and businesses, how might this reshape payment journeys and customer expectations?

I think we need to distinguish agentic commerce from agentic payments.

Commerce will probably become agentic first. Discovery and comparison are shifting from search bars and merchant sites into AI interfaces, and merchants who are not machine-readable will simply get bypassed before checkout is ever reached.

Payment execution follows later, with added guardrails. Within payments, there is a clear ladder. The agent recommends a method and the consumer pays; then, the agent stages the basket and the consumer approves it. Eventually, it transacts inside guardrails set once — amount, merchant, category, timing. The first use cases are repetitive, lower-risk purchases: subscriptions, groceries, utilities, household consumables, prescription refills… In higher-consideration categories such as travel, electronics or luxury, AI will initially have more influence over discovery than over the final payment execution.

The expectation shift is the interesting part. Today, a customer asks, “Did I get a good deal?” Tomorrow, they will assume their agent already optimised it across price, fees, rewards and protection. Approval collapses from every transaction into a mandate granted once.

And the winning payment method becomes the one the agent can read, trust and execute — a very different competition than the one issuers have been running. Brand and loyalty will still matter, but they will need to translate into machine-readable value at the moment of purchase.

Trust is the foundation of any payment ecosystem. How should financial institutions and payment providers balance innovation with security, transparency, and regulatory compliance in an AI-driven environment?

Our consumer research shows exactly where the line sits today. Trust in AI is high for research purposes but falls sharply the moment money moves: only about a third of US consumers trust AI to handle their payment details securely or would let an agent transact without confirming first.

I would not read that as resistance to AI — most people have never had an agentic payment experience yet, whether good or bad. The issue is that trust today is inferred from the consumer being present and actively approving the transaction. In an agentic environment, that implicit signal has to be replaced by explicit evidence: verified agent identity, human-to-agent binding, scoped and revocable credentials, audit trails, and clear liability when something goes wrong. That infrastructure does not yet exist – not yet as a common standard.

Europe could have some real advantages here — instant payments are live and mandated, and verification of payee is in production. The Payments Service Directive 3 (PSD3) and the Payments Server Regulation (PSR) are expected to apply around 2028; nonetheless, it is worth noting that the framework was designed for human-authorised transactions, and nobody has yet answered who bears the loss when an agent exceeds its mandate.

My advice is not to wait for the rulebook. Payment providers and networks should build controls now, engage regulators while the defaults are still being written, and treat trust as a product rather than a compliance cost. It is one of the few genuinely monetisable layers in this stack, and today it has many contenders, but no clear winner.

Payment providers are facing multiple disruptive forces simultaneously — real-time payments, digital assets, embedded finance, AI. Which of these trends do you believe will be the most transformative, and why?

AI — but not because AI replaces all the other trends; quite the opposite: AI can become the intelligence layer sitting above them.

Real-time payments, stablecoins, and embedded finance change how money moves – they create new rails, introduce potentially programmable forms of money and settlement, and put payments inside other customer journeys. But an AI agent can sit above all those solutions and decide: “which one should I use now?”

A platform agent may care about relevance, merchant participation and its own economics. A bank or wallet agent may optimise for the consumer: rewards, budget, credit line, purchase protection and trust. A merchant-controlled agent optimises for acceptance cost and conversion — and that is what could finally give a low-cost rail a reason to be chosen, because a human at checkout has no incentive to think about the most convenient acceptance cost for the merchant.

The underlying inputs may be similar, but the weighting changes depending on who owns the agent. This is why I think one of the fundamental questions for the industry becomes: who owns the decision layer?

What new opportunities does AI create for banks, payment providers, and fintechs beyond improving operational efficiency and fraud management?

AI creates four main opportunities beyond operational efficiency and fraud management, all linked to agentic commerce and agentic payments; and it is worth noting they are growth opportunities, rather than cost ones.

First, AI creates the opportunity to become the consumer's agent. Wallets and issuers already hold the credential, the spending data and the trust relationship. They are well placed to be the agent — but only if they move before the large platforms establish the habit, and consumer comfort today sits with technology brands rather than bank apps.

Second is about selling trust. Agent verification, mandate validation, agent-aware authorisation, dispute rails adapted to agent-initiated transactions — these are products, not overheads, and networks and banks are the natural owners.

The third involves orchestration. Acquirers and payments service providers (PSPs) can move from passive processing to active routing across cards, account-to-account payments, instant rails and stablecoins, offering merchants one connection point across a fragmented set of agent protocols.

Finally, AI will allow us to rethink loyalty. If rewards no longer influence a human directly, they have to be expressed as structured logic that an agent can read and optimise against. That is a redesign of the product, not a marketing refresh.

And there is a demand-side benefit people tend to underestimate: AI-generated recommendations convert several times better than traditional search. Handled well, this is a sales channel, not just a margin threat.

From your work with financial institutions and payment companies across Europe and globally, how prepared is the industry today for the shift toward AI-powered commerce and decision-making?

Awareness is high; readiness is low. Almost every institution I work with has an AI program, but very few have an agentic commerce answer — meaning a quantified view of which volumes are exposed, whether their products and data are machine-readable, and what happens to the Profit & Loss (P&L) if 10%-30% of e-commerce volume becomes agent-routed.

The infrastructure race, by contrast, is well underway. The networks have launched agent credentials and trusted-agent protocols, the large PSPs have shipped agent-facing Application Programming Interfaces (APIs), and more than ten protocols are competing to define how agents and merchants talk.

The more difficult gaps are fragmented data, legacy journeys, unclear ownership of the customer relationship, limited machine-readable product information, and unresolved questions about liability and recourse – which all point to a lack of merchant readiness and consumer trust.

Looking ahead, what do you think will define the next generation of payments, and what should industry leaders be prioritising now to remain competitive?

The defining shift is from being familiar to humans to being legible to machines.  To get to that stage, I would prioritise three things.

The first is to size the exposure honestly — how much of your volume is habitual, low-consideration spend, and what the P&L looks like at meaningful agent-routed share.

Second, decide your role — own the agent, own a control point in demand conversion, execution or authorisation, or be preferred infrastructure — and be honest about which is realistic.

And lastly, set explicit triggers for strategic action: protocol convergence, merchant agent-readiness, delegated spend crossing double digits of e-commerce volume, and regulatory clarity.

Our view is that agentic commerce could represent somewhere between 15%-25% of US ecommerce by 2030. Europe will move later and differently. But defaults, once set, are extraordinarily hard to unseat —that window is open now, but it will not remain open indefinitely.

About Mariagiovanna di Feo

Mariagiovanna Di Feo is a Partner in Bain & Company's EMEA Financial Services and global Payments practices. With over 15 years of experience since joining Bain in 2010, she advises leading financial institutions on strategy, M&A, post-merger integration, and customer excellence. Her expertise spans payments, retail ecosystems, and fintech, helping clients evolve business models and drive growth. Mariagiovanna holds a master's degree in economics, cum laude, from Bocconi University.

 

About Bain & Company 

Bain & Company has partnered with leaders worldwide since 1973 to solve their toughest challenges and deliver enduring results. The firm works with clients including private equity and portfolio companies to build capabilities, stay ahead of change, and redefine industries. Bain measures its success by its clients' success, holding the highest client advocacy levels in the field. Recognised as one of the best places to work, they operate as one global team of strategists, experts, and technologists.