Two Sides of AI Shopping Agent Checkout

At Google I/O event this past week, it was announced that Google Shopping’s agentic checkout will be available in a few months. Similarly, quite a few companies launched or limited launched agentic checkout features including Amazon’s “Buy for Me” and Perplexity’s “Buy with Pro”. The age of AI shopping agents is coming. A big piece of AI shopping agents is the checkout. Before the Google I/O event, quite a few card networks and payment processor companies including Visa (Intelligent Commerce), Mastercard (Agent Pay ), PayPal, and Stripe, announced the agentic payment innovation.

BorderX Lab’s CloudStore AI product has been offering the agentic checkout API to our external clients for more than two years. This API has been used internally even longer to support shopping agents Beyond app, a cross-border shopping assistant for Chinese consumers, and BeyondStyle, a price-discovery focused shopping assistant for American consumers, over a decade. This agentic checkout has successfully fulfilled $1,000,000,000 worth merchandising items such as Nike Sneakers and Lululemon yoga pants, and delivered 5,000,000 orders around the world. So CloudStore AI checkout API is a battle tested tool for AI shopping agents. We have accumulated a lot of operating experience with agentic checkout. Here we like to emphasize there are two sides of AI shopping agent checkout. One side is the communication between humans/apps and the shopping agents. The other side is the communication between AI shopping agents and merchants. We observe that the communication between AI shopping agents and merchants are underdiscussed. We want to emphasize that both sides are equally important and affect the overall shopping experience greatly. On each side, we are dealing with different challenges.

One side is between humans/apps and the shopping agents. Recently many card network companies (e.g., Visa and Mastercard) and payment companies (e.g., Stripe and PayPal) announced agentic payment solutions for AI shopping agents. They are mostly focused on the security and trust between consumers/apps and AI shopping agents. For example, how much money will a consumer authorize an AI shopping agent to spend during agentic checkout. Based on our many years’ operating experience, we have served this side pretty well. Our AI shopping agents are first-party AI shopping agents. We have fine-tuned authorization rules about AI shopping agents for a long time. For example, we optimize how much headroom we assign to AI shopping agents to deal with the unexpected price change on the merchant side. For third-party AI shopping agents, consumers/apps could not fully trust third-party AI shopping agents, at least in the beginning. Consumers have to clearly authorize AI shopping agents to do tasks with risk control in mind.

The other side is between AI shopping agents and merchants like Walmart. This is also a very important side of AI shopping agent checkout process. It is related to payment success, fulfillment success, logistics tracking, and many more. It greatly affects consumer’s shopping experience. For example, many merchants are deploying many mechanisms including reCAPTCHA to block orders from agents. If the AI shopping agents could not fulfill the orders at a high success rate, you can imagine how an AI shopping agent and a consumer will feel. Based on our many years of operating experience, we summarize two key metrics to consider when we evaluate the collaboration between AI shopping agents and merchants.

Payment Success Rate, i.e., what is the percentage of payment actions initiated by AI shopping agents that succeeds. AI shopping agents may fail to pay for many possible reasons including reCAPTCHA verficcation failure and IP address blocking.
Fulfillment Success Rate, i.e., what is the percentage of orders successfully made by AI shopping agents that ships out. Merchants may cancel orders for many possible reasons such as out of inventory.

Besides the payment collaboration between AI shopping agents and merchants, AI shopping agents will also track the delivery time, merchandising item quality, and even reverse logistics. We believe that this side of AI shopping agent checkout is under-discussed while it is extremely important to drive the AI shopping agent’s wide adoption.