Agentic Commerce Will Be Won in the Last Click
PayPal’s UK agentic-commerce research, published on 5 August, is another indication that retail’s next interface may be conversational. A shopper can ask an assistant to find a product, compare specifications or narrow a confusing choice set before visiting a retailer’s site.
That possibility is commercially significant. It does not, however, mean consumers are ready to delegate purchasing wholesale.
The more useful distinction is between assisted discovery and delegated commitment. Brands that collapse the two will over-invest in novelty and under-invest in the moments where trust, money and accountability meet: product claims, price, delivery, returns and payment.
Discovery is moving, decision rights are not
The immediate impact of AI shopping tools is likely to be felt near the top of the funnel. Search, comparison and recommendation are effort-heavy tasks, particularly in categories with dense specifications, fragmented reviews or too much apparent choice. A competent assistant can reduce that effort.
But reducing search effort is different from removing the need to decide. Ipsos’s latest *Shopping With AI* work, featured again in its August update, finds that 27% of AI-aware consumers already use such tools for product research, while only 9% currently allow an AI tool to make autonomous purchases. Just under a quarter say they would even consider that level of delegation. Ipsos also reports that 71% believe AI should be approached cautiously.
The commercial opportunity is not autonomous shopping at any price; it is confidence at the point where a recommendation becomes a transaction.
This creates a more demanding design brief than simply making a catalogue accessible to an assistant. Retailers need to decide which tasks customers can safely hand over, which require confirmation, and how the customer can inspect or reverse an automated choice.
A repeat purchase of a known household item may be suited to a high degree of delegation. A first-time purchase involving fit, health, a major financial commitment or a subjective aesthetic judgement plainly is not. The same customer can hold radically different thresholds across categories.
The new research question is permission
The conventional e-commerce funnel records impressions, clicks, conversion and basket value. Those measures remain valuable, but agentic commerce adds a more revealing question: what authority did the customer give the system at each stage?
That means measuring more than whether an AI assistant appeared in a journey. Businesses should distinguish between customers who use one to generate ideas, compare options, build a basket, initiate payment or complete a purchase without reviewing the order. Each action signals a different level of trust and a different operational risk.
The measurement challenge extends beyond the retailer’s own interface. If product discovery increasingly happens in third-party assistants, a brand may see less of the consideration journey. Its response should not be to treat every recommendation as a black box. It should test whether product data is clear enough to be represented accurately: dimensions, compatibility, stock status, delivery restrictions, warranties, ingredients, pricing conditions and return rules.
In practical terms, the product-information function becomes part of customer acquisition. Incomplete or ambiguous data does not merely impair a webpage; it raises the chance that an assistant makes a weak recommendation, or that a customer encounters a contradiction after clicking through.
The trust context is not abstract. The Office for National Statistics reported in July that 77% of adults were concerned about personal data being used without consent, while 63% identified increased cybercrime as a possible negative impact of AI. For retail teams, these concerns should shape consent journeys and payment design, rather than be relegated to legal copy.
Checkout becomes a proof point
The final click is where agentic commerce encounters the obligations of ordinary commerce. A recommendation must translate into an accurate order; a price must be intelligible; a delivery promise must hold; customer support must be reachable when something goes wrong.
This is why the operational foundations matter more than an impressive demonstration. Retailers should be able to show customers what an assistant has selected and why, before money moves. They should give clear controls over substitutions, price ceilings, preferred brands, data sharing and order approval. For recurring purchases, those settings need to be easy to find and amend.
Returns deserve equal attention. Delegated buying could increase conversion for routine purchases, but it could also create costly dissatisfaction if the selected product is unsuitable or the customer does not understand how the choice was reached. Return reason codes should therefore distinguish between a customer’s own selection, an AI-suggested item and an automated order. Without that distinction, retailers will misdiagnose whether a problem sits in assortment, information quality, recommendation logic or fulfilment.
This is also a reason to resist a narrow automation narrative. The ONS reports that 29% of UK businesses used at least one AI technology in June, with text generation and visual-content creation the most commonly reported applications. ONS business analysis suggests adoption is already broadening; it does not establish that customers have granted equal permission for businesses to automate consequential decisions.
Build for graduated delegation
The strongest commercial strategy is likely to be incremental. Start with useful, low-risk assistance: guided comparison, transparent recommendations, replenishment prompts and better service triage. Then test where customers opt into greater autonomy, by category and customer segment, while preserving an obvious human override.
The crucial unit of analysis will be neither the chatbot session nor the headline conversion rate. It will be the delegation threshold: the point at which a customer moves from accepting help to surrendering control.
Retailers that identify that threshold accurately can remove friction without creating unease. Those that assume convenience automatically equals consent may discover that the most valuable part of the purchase journey was never search. It was reassurance.
Comments