AI brokers are shopping for on our behalf. Xavier Sheikrojan, director of threat intelligence at Signifyd explains how agentic commerce shifts cost alternative, SCA, fraud alerts and legal responsibility—and what retailers should do subsequent.
A purchase order is made. The product is chosen, the cost authorised and a parcel is dispatched. They didn’t even browse. In some circumstances, they might not have even identified the transaction occurred till after the very fact. This isn’t the begin to an inconceivable riddle. It’s agentic commerce – the place AI brokers act on behalf of the buyer, initiating and finishing purchases throughout the net with none directinvolvement from the person.
Amazon’s “Purchase for Me” affords an early glimpse, permitting customers to buy with third-party retailers completely throughout the Amazon app. The agent selects the merchandise, fills within the kinds and completes the cost with out the consumer ever leaving Amazon’s ecosystem. For the service provider, thecheckout course of is diminished to an off-stage operate.
This implies retailers are being written out of their very own transaction circulation. With no management over how their product is introduced, how cost is dealt with or what safeguards are utilized, the dangers compound rapidly: if the buyer isn’t initiating the transaction, who is selecting the cost technique? Who’s authenticating it? And if one thing goes fallacious, who’s liable?
For now, these agent led flows stay comparatively new. However as conventional checkout processes and fraud checkpoints are bypassed, the basics of how funds are chosen, authenticated and guarded might want to change.
AI brokers are the brand new cost resolution makers
In agentic commerce, the selection of cost technique is now not a point-of-sale resolution made by customers. It’s a step carried out by the agent, usually earlier than the buyer even is aware of the transaction is underway. Performing as a cost adviser, the agent evaluates choices: accessible loyalty balances, probability of issuer approval, processing pace, value to the service provider or evenincentives negotiated between platforms and cost suppliers.
This repositions the decision-making course of. The locus of management shifts upstream to the agent, leaving the retailer with far much less affect over the cost combine than they at present have. As soon as a purchase order is handed into the agent’s surroundings, the service provider’s checkout flows andpromotional prompts now not form the result.
This may speed up the adoption of sure strategies. BNPL and instalment choices will be surfaced precisely when they’re almost definitely to transform. Digital wallets, with their tokenised credentials and near-instant authentication, are properly suited to agent-led transactions and couldbecome the default in lots of eventualities. Strategies that introduce friction, carry greater prices, or lack API-driven integration might discover themselves deprioritised by default.
For retailers and cost suppliers, the influence is structural. A unique cost combine impacts processing prices, settlement cycles and publicity to fraud. The service provider may not even see the alerts that when knowledgeable their fraud checks if the transaction is initiated and accomplished throughout the agent’s area.
And as soon as cost choice sits with the agent, one other query follows: who has the authority and obligation to authenticate the cost?
Regulatory uncertainty
PSD2’s Sturdy Buyer Authentication (SCA) frameworks have been constructed on the belief that the buyer is the one initiating the cost. Agent-led flows problem that premise. An AI agent making the acquisition on a consumer’s behalf doesn’t map cleanly to “merchant-initiated transactions” and but just isn’t strictly a customer-initiated cost both.
Delegated authentication is feasible, however solely with formal issuer agreements and a transparent assumption of legal responsibility by the platform. With out this, ambiguity emerges over who applies SCA which dangers failed challenges, issuer declines and finally non-compliance.
This uncertainty just isn’t confined to Europe. Increasing 3DS necessities in markets like Japan and Australia are elevating comparable questions on learn how to apply authentication with out a human current. Offsite, agent-led flows additionally make exemption possession unclear. Whether or not theexemption is predicated on transaction threat evaluation, low-risk thresholds or whitelisting, it’s not at all times apparent who’s chargeable for making use of it. That lack of readability can add friction and cut back conversion charges.
In EMEA, the stakes are notably excessive. Retailers might want to rethink exemption methods and authentication flows to make sure agent-led funds don’t erode conversion. And the success of agentic commerce right here might rely as a lot on shopper notion as onregulatory readability.
European customers are typically privacy-conscious, security-aware and accustomed to sturdy identification checks. The thought of an AI making purchases on their behalf might really feel intrusive. With out transparency, clear consent flows and a powerful belief framework, retailers threat dropping buyer confidence earlier than the expertise has an opportunity to scale.
Disputes in the dead of night
Agent-led funds introduce new dispute eventualities that in the present day’s frameworks will not be ready to deal with. A buyer may declare “I didn’t authorise the agent to make use of this cost technique” or “I by no means agreed to that buy being made this fashion.” Untangling these claims is more durable when the order arrives with much less context than a typical eCommerce transaction.
When the acquisition is accomplished throughout the agent’s surroundings, retailers might lose visibility of the alerts they depend on for fraud detection. In some circumstances, even the service provider’s title might not seem clearly on the client’s financial institution assertion, rising the danger of disputes and model confusion. Fraudsters are already experimenting with methods to imitate trusted agent behaviour to slide previous detection programs, and most present fraud instruments will not be but educated to identify the nuances of AI-driven exercise.
Conventional defences have been designed for human behaviour. They usually wrestle to detect the fast, non-human checkout patterns of an AI agent. With out diversifications, retailers might discover themselves unable to tell apart between a professional agent-led transaction and a fraudulent one.
Legal responsibility additionally turns into extra complicated. Totally different cost varieties comply with totally different dispute guidelines. As an illustration, a BNPL dispute is dealt with otherwise from a card chargeback, and if the cost is processed offsite, the service provider might not be capable to apply their very own fraud controls in any respect. That lack of visibility and management can depart them uncovered to greater chargeback charges, weaker proof in disputes, and finally diminished safety for each income and model fame.
Agentic commerce adjustments the place worth and threat sit within the funds chain. Choices as soon as made on the level of sale are shifting to upstream environments managed by platforms, that means that affect over cost alternative, authentication and legal responsibility will consolidate in fewer fingers.
Quite than focusing solely on optimising their very own checkouts, retailers might want to place themselves contained in the agent’s resolution logic by way of information sharing, fraud sign integration or industrial incentives. Those that safe that place might acquire most well-liked routing and higher conversion; those that don’t threat being deprioritised completely.
The subsequent part for the funds business is to determine shared guidelines for consent, authentication and dispute dealing with in agent-led flows. With out them, belief and efficiency will erode. With them, funds might change into sooner, safer and higher matched to shopper wants.
Xavier Sheikrojan is a seasoned skilled in threat administration and fraud prevention, at present serving because the Senior Threat Intelligence Supervisor at Signifyd since September 2021, the place obligations embrace constructing and managing the Threat Intelligence staff and implementing revolutionary threat methods. Previous to this function, Xavier was a Fraud Supervisor at Glovo, main a high-performing fraud staff and contributing to fraud coverage growth.
Source link
