Agentic ecommerce: What it is and what it means for retail businesses

AI is everywhere, and people use it in all kinds of ways. Some use it to simplify complex spreadsheets, some ask it for dinner recipes, some treat it as a sounding board, and some use it to compare products. This latter category is rapidly evolving into what the industry calls agentic ecommerce, which is the focus of this article.

In agentic ecommerce, an artificial intelligence agent does not just suggest an item; it actively evaluates options and can complete the purchase on the user's behalf. Instead of a shopper spending an hour reading reviews to find the best marathon running shoes for flat terrain under €120, the software runs the search, compares the specifications, and presents the single best option ready to buy. It sounds like the ultimate convenience, but the reality of letting a machine take over the shopping cart is proving to be far more complicated.

Why shoppers and systems struggle with automated buying

It remains to be seen if consumers actually want to hand over the checkout process to a machine. While automated buying makes perfect sense for restocking household necessities like laundry detergent or coffee beans, other purchases naturally require more human involvement.

For many, online shopping is a form of leisure. It is the digital equivalent of window shopping. People enjoy scrolling through different styles, reading reviews, and discovering unexpected items. Giving up that discovery process to a machine fundamentally conflicts with how people shop for pleasure.

Beyond the loss of discovery, there is a significant trust barrier when real money is on the line. Shoppers know that artificial intelligence can make mistakes or misinterpret prompts. When deciding to buy, people naturally hesitate to relinquish control. They fear the software might purchase the wrong size, ignore a crucial shipping fee, or select a slightly different model than what they actually wanted.

The systems themselves also struggle to provide a reliable shopping experience. A recent study by SparkToro revealed that these tools are highly inconsistent when recommending brands or products. The research found there is less than a 1 in 100 chance that ChatGPT or Google's AI will give a user the same list of brands in any two responses across 100 runs. Furthermore, there is only a 1 in 1,000 chance of seeing two lists generated in the exact same order.

To see this in action, we ran our own test. We asked ChatGPT the exact question from our earlier example four times in a row: "What are the five best marathon running shoes for flat terrain under €120?"

As the image shows, the results were inconsistent at best. In the first run, the system picked the Pegasus 41 as the single best shoe for everything. In the second run, it chose the Revel 8, and Pegasus didn’t even get a mention.

If you ask a system to find a specific product and receive a different answer almost every single time, the recommendation loses all value. It means the software is not actually evaluating which shoe is objectively the best based on real data. Instead, it is just guessing and generating a list that sounds plausible. If the answer changes constantly, the response is inherently unreliable. This defeats the entire purpose of delegating the decision to an expert system in the first place.

What’s more, this inconsistency makes it nearly impossible for businesses to reliably market their products. A retailer cannot simply optimise a product description and expect the software to consistently recommend their shoes, because the algorithm's choice changes randomly with every prompt.

How the financial sector is securing automated checkouts

If consumers are going to trust software to spend their money, the entire system needs a massive upgrade in reliability. People will not accept random guesses when their bank accounts are involved. Interestingly, the blueprint for solving this trust issue is already being drawn up by the banking industry.

Banks are currently building the foundation for automated transactions, but they are taking a very different approach from the tech companies. Instead of aiming for fully independent software, the financial sector is building systems based on strict human oversight.

In this model, the software acts as an assistant rather than an independent buyer. For example, a bank customer might ask the software to handle a disputed charge. The system gathers the necessary forms, checks the policies, and drafts the messages, but it stops there. The human customer reviews the work and clicks the final confirmation button.

Research shows that consumers strongly prefer this cautious approach. According to the Accenture Banking Top Trends 2026 Report, 82% of banking clients want to approve every action taken by an automated agent.

To make these checkouts a reality, major payment networks are actively building new security rules. Visa is developing a trusted agent protocol. This system ensures the software agent is legitimate and sets strict limits on what the software is allowed to do. Mastercard has a similar project called Agent Pay. It requires agents to be formally registered and uses tracking codes so every single payment can be traced back to the source.

Payment processors are also preparing for this shift. Stripe is developing the underlying technology to enable software to process payments securely, and Klarna plans to use this same technology to offer its payment options during automated checkouts.

By putting human approval first and tracking every move the software makes, the financial sector is creating the exact guardrails that ecommerce needs.

Why retailers want to keep control of the final sale

Even if payment networks make automated transactions perfectly secure, giving up the final sale to an external software platform comes with massive risks. Retailers are discovering that they cannot simply hand over their virtual shop floor to a general-purpose artificial intelligence.

Walmart provides a clear example of this lesson. Initially, the retail giant partnered with OpenAI to let customers buy items directly inside ChatGPT using a tool called Instant Checkout. The goal was ultimate convenience, allowing the software to handle the entire shopping trip from product search to payment.

However, the experiment failed to generate the expected sales. According to Search Engine Land, the conversion rates for purchases made directly inside the chat interface were three times lower than when users clicked a link and completed the purchase on Walmart's actual website.

The core problem was a lack of direct integration. The third-party software struggled to reliably verify exact stock levels and promise accurate delivery times. When shoppers cannot trust that an item is actually in stock, or they do not know exactly when it will arrive, they naturally abandon the purchase.

To regain control of the customer experience, Walmart ended the exclusive partnership. As reported by Retail Dive, the retailer instead built its own software agent named Sparky and embedded it inside the chat platforms. By doing this, Walmart kept absolute control over its inventory data, pricing, and the final checkout process.

Following this shift, OpenAI changed its approach to focus on product discovery rather than processing the final transaction. This situation proves a crucial point for the future of agentic ecommerce. Major retailers are unwilling to surrender their customer data and the final checkout experience to general technology companies, no matter how advanced the software might be.

Practical ways smaller retailers can use artificial intelligence today

It is completely understandable if the thought of adding more artificial intelligence to your business sounds exhausting. The industry hype is deafening, but actual willingness to pay for these tools tells a different story.

Consider Microsoft Copilot: despite being integrated into the world's most ubiquitous office software and aggressively pushed on every user, recent FY26 Q2 data shows only 15 million paid Copilot seats. Significant in isolation, but set against more than 450 million Microsoft 365 commercial seats, it represents a measly 3.3%.

As the chart illustrates, while the push for AI is everywhere, the overwhelming majority of businesses are not yet convinced to open their wallets for general-purpose assistants. It is very tempting to tune it out entirely.

However, ignoring the shift entirely means missing out on practical, highly targeted tools that actually drive sales right now. While the industry waits for fully automated checkouts to become mainstream, there are smart ways to use artificial intelligence to increase revenue without frustrating buyers or overhauling your entire operation.

A strong example is product recommendations. Instead of showing customers a generic list of trending items, AI can be used to match and display related products that other people actually bought alongside their specific order. It provides a highly relevant recommendation based on real, verifiable shopping patterns.

Looking further ahead, small and medium retailers do not need to build complex custom software like Walmart did to prepare for true agentic ecommerce. Major ecommerce platforms are already building the technical infrastructure in the background.

For example, Shopify recently launched the Universal Commerce Protocol. This open standard allows external software agents to natively read merchant catalogues and process payments. The platform handles the communication automatically, meaning merchants do not need to hire developers or arrange complex integration meetings. To prepare for the eventual wave of automated shoppers, smaller businesses simply need to keep their product data accurate and well-organised so the software can easily read it. The businesses best placed for that shift are the ones who start now.