Somewhere between a consumer saying 'book me the cheapest flight to Edinburgh next Thursday' and a confirmation landing in their inbox, a transaction is completed — not by a human, but by software acting on their behalf. This is agentic commerce, and it is no longer a speculative concept. Both Visa and Mastercard have begun rolling out infrastructure that allows AI agents to hold tokenised payment credentials and execute purchases autonomously, without a human approving each step. For UK ecommerce businesses still treating AI as a customer-service add-on, this shift represents something far more structural: a fundamental change in who — or what — is actually buying from you.
The commercial implications are significant and immediate. If an AI agent is the entity completing a purchase, your conversion funnel, your affiliate attribution model, your fraud detection logic, and even your returns policy all need rethinking. The question facing senior decision-makers is not whether agentic transactions will reach meaningful volume — they will — but whether your commercial infrastructure is built to handle them when they do.
What Agentic Commerce Actually Means
An AI agent, in this context, is a software system that can perceive goals, reason about options, take actions, and complete multi-step tasks without human input at each stage. Applied to commerce, this means an agent can receive a standing instruction — 'reorder my dog food when stocks fall below two bags' or 'find and book a hotel under £150 per night for my Liverpool trip' — and execute the full purchase lifecycle: searching, comparing, selecting, and paying. The payment step is what has historically been the barrier. You cannot complete a purchase without payment credentials, and until recently, those credentials lived with the human.
Visa's Agent Pay programme and Mastercard's Agent Pay initiative (both announced and iterated upon through 2025) address this directly by allowing tokenised credentials to be issued to, and controlled by, AI agents operating within defined spend parameters. These are not stored card details in the traditional sense — they are scoped tokens with configurable limits on merchant category, spend ceiling, and validity period. The consumer delegates authority; the agent executes within it. For merchants, the practical effect is that a confirmed order may arrive with no human having actively chosen your product in the moment of purchase.
The Affiliate Attribution Problem
Affiliate and referral marketing is built on a human journey: a person reads a review, clicks a tracked link, browses a product page, and converts. Commission is attributed to the publisher or partner who influenced that journey. Agentic commerce disrupts every assumption in that model. When an AI agent selects a product, there may be no click, no referral link traversal, and no identifiable moment of human influence to attribute. The agent may be acting on instructions set weeks earlier, drawing on a product feed it ingested via API, or optimising against a set of rules with no relationship to any affiliate content.
For UK businesses running affiliate programmes through networks such as Awin, Rakuten, or in-house solutions, this creates a near-term structural gap. If a meaningful share of transactions begins arriving via agentic pathways — particularly in high-frequency categories like grocery, consumables, travel, and B2B procurement — the existing attribution infrastructure will systematically under-report partner contribution and over-credit direct. Worse, without proactive changes, businesses risk paying no commission on sales that were genuinely influenced by a partner's content or data feed, damaging publisher relationships and distorting optimisation decisions.
Restructuring Incentives for Software Buyers
The practical response is to treat the AI agent as a distinct customer segment in your commercial architecture — not a quirk to route around, but a channel to design for. This starts with your product data. Agents do not respond to lifestyle imagery or persuasive copywriting; they parse structured data. Investing in clean, machine-readable product feeds — accurate attributes, reliable stock signals, consistent pricing — becomes a direct competitive advantage when agents are selecting suppliers algorithmically. A poorly structured product catalogue will simply be excluded from an agent's consideration set before a human ever sees it.
On the affiliate side, the answer lies in shifting from click-based attribution to feed-based or API-based partnership models. If a price comparison service or product recommendation engine is surfacing your products to agents, the commercial relationship should reflect that — through data licensing arrangements, cost-per-acquisition deals negotiated at the feed level, or structured integration fees. This is already how B2B procurement platforms operate, and consumer-facing agentic commerce is converging on similar models. Businesses that adapt their partner incentive structures now will be better positioned to attract the AI platforms and agent ecosystems that will increasingly control purchase intent.
Fraud, Identity, and the Trust Infrastructure
Tokenised agent credentials introduce new fraud surface areas that UK merchants and their payment providers need to address proactively. Traditional fraud signals rely heavily on behavioural biometrics — mouse movement patterns, session duration, typing cadence — that are irrelevant when the purchaser is software. An agent executing a transaction will look, to a legacy fraud model, like a suspiciously efficient human: no browsing hesitation, precise form completion, no cart abandonment. Fraud rules calibrated on human behaviour will generate false positives and false negatives at scale.
The mitigation strategy here involves working with your payment processor to understand how agent tokens are flagged and verified at the network level, and ensuring your own fraud and risk tooling can ingest those signals. Equally important is the question of liability in disputed transactions: if an agent purchases on behalf of a consumer who later claims they did not authorise the specific transaction, the chargeback implications for merchants are not yet uniformly settled across UK card scheme rules. Engaging your acquiring bank and legal counsel on this now — before volume makes it urgent — is prudent risk management.
Agentic commerce is not arriving all at once, and most UK ecommerce businesses have a meaningful window to adapt rather than react. But that window is measured in months, not years. The practical starting points are straightforward: audit your product data quality through the lens of machine readability, map which of your affiliate partnerships would survive the removal of click-based attribution, and open a conversation with your payment provider about agent token handling and fraud signal changes.
The businesses that will perform well in an agentic commerce environment are those that treat structured data, API accessibility, and feed-based commercial relationships as strategic investments rather than technical housekeeping. At iCentric, we work with UK organisations to build the commercial and technical infrastructure needed for these emerging transaction models — from product feed architecture to partner programme restructuring. If your current setup was designed for human browsers, now is the right time to examine what needs to change.
Are Visa and Mastercard's agent payment schemes available to UK merchants now, or still in pilot?
Both Visa's Agent Pay and Mastercard's equivalent programme were announced and iteratively developed through 2025, but merchant-facing availability depends on your acquiring bank and payment processor having integrated the relevant APIs. UK merchants should contact their payment provider directly to understand current availability and any onboarding requirements. Rollout is gradual and category-dependent.
How do tokenised agent credentials differ from a saved card or a digital wallet like Apple Pay?
Unlike a saved card or wallet, agent tokens are scoped credentials issued specifically for automated use — they carry configurable constraints such as spend limits, permitted merchant categories, and expiry windows set by the cardholder. The key difference is that the token is designed to be operated by software autonomously, rather than presented by a human at the point of purchase. This scoping is intended to limit risk if the agent acts outside its intended parameters.
Will agentic transactions be subject to Strong Customer Authentication under UK regulations?
This is an evolving regulatory question. SCA rules under the UK's retained version of PSD2 were designed around human-initiated transactions. The Financial Conduct Authority has not yet issued definitive guidance on how SCA applies to fully autonomous agent-initiated payments. The expectation is that the initial delegation of authority — when the consumer sets up the agent's payment scope — will carry the authentication event, but merchants should monitor FCA publications and take legal advice for high-value or high-risk categories.
What product data formats should we prioritise to make our catalogue agent-readable?
Structured data using Schema.org Product markup, clean and complete Google Merchant Centre-compatible feeds, and well-documented REST or GraphQL APIs are the most immediately relevant formats. Accuracy of stock availability, consistent attribute naming, and machine-parseable pricing (including VAT status) are more important to agents than rich media or descriptive prose. An audit against these criteria is a practical first step.
How should we handle returns and refunds when the purchaser is an AI agent rather than a human?
Your returns policy terms almost certainly reference 'the customer' in ways that assume human decision-making and consumer rights entitlements. Legal advice is warranted on whether a consumer's statutory rights under the Consumer Rights Act 2015 apply when a delegated agent made the purchase, and how disputes should be documented and evidenced. Practically, you should ensure your order management system can flag agent-originated transactions so customer service teams handle them with appropriate context.
Could AI agents be gamed by competitors or bad actors to generate fraudulent orders?
Yes — agent impersonation and credential abuse are genuine threat vectors. Attackers could attempt to spoof legitimate agent tokens or exploit misconfigured spend scopes. Merchants should work with their fraud platform providers to develop detection rules specific to agent transaction patterns, and ensure that any unusually high-velocity order sequences from what appear to be agent sessions are flagged for manual review. Network-level token verification from Visa or Mastercard provides a baseline, but it is not sufficient on its own.
How do we value a partner's contribution to a sale if there is no affiliate click to track?
The shift is towards contribution-based or data-based commercial models. If a partner's product feed, API, or recommendation engine surfaces your product to an agent, that contribution can be measured through feed impression data, API call logs, or negotiated on a cost-per-acquisition basis tied to order-level identifiers passed at checkout. This is analogous to how B2B procurement integrations are currently monetised, and it requires updated commercial agreements with partners rather than relying on cookie-based attribution networks.
Are there specific ecommerce sectors where agentic commerce will accelerate fastest in the UK?
The earliest meaningful volumes are likely in categories with high repurchase frequency and low selection complexity: grocery and household consumables, travel and accommodation booking, B2B office supplies and MRO procurement, and utilities switching. These are domains where standing instructions are easy to define and where price and availability are the primary selection criteria — exactly the conditions that favour algorithmic purchasing over human browsing.
Should we update our terms and conditions to explicitly address AI agent transactions?
Yes, and relatively urgently. Your current terms and conditions almost certainly do not contemplate a non-human purchaser, which creates ambiguity around contract formation, liability, and dispute resolution. At minimum, you should add clauses addressing delegated authority, the relationship between the agent operator and the underlying cardholder, and how disputes will be handled when an agent acts outside its intended scope. A commercial solicitor with fintech or ecommerce experience is best placed to draft appropriate language.
How should a technical lead prioritise internal development work to prepare for agentic commerce?
The highest-impact technical investments are: ensuring your product catalogue is exposed via a clean, well-documented API or feed; adding order-source metadata fields to your order management system so agent-originated transactions can be identified and routed appropriately; and reviewing your fraud scoring pipeline with your payment provider to understand where current rules will need adjustment. These are relatively contained changes that deliver value independently of agentic commerce volume and build the foundation for more sophisticated agent integrations later.
Get in touch today
Book a call at a time to suit you, or fill out our enquiry form or get in touch using the contact details below