Key insights: Tech giants such as Google and OpenAI are pushing to bring checkout into their platforms, but for many merchants, solving the question of payment inside the LLM is less of a priority.What’s at stake: Merchants are a necessary cog in enabling an agentic commerce ecosystem. Forward look: Retailers this year will be prioritizing data integrity and infrastructure to ensure that their product catalogue is discoverable by AI agents. 

Big technology firms are pushing agentic commerce, and the tech is a winner with retailers. But merchants are more interested in attracting shoppers than embracing bot payments. 

Processing Content

Large language model providers such as Google, OpenAI, Anthropic and Perplexity are all pushing to bring the entire purchase experience – from discovery to payment – into their ecosystems. 

Retailers, especially larger companies, recognize the importance of artificial intelligence and agentic commerce, and plan to make that a reality this year, Srjana Balraj, global head, unified commerce platform at TCS told American Banker. 

“If you are not visible in the [AI] chat interface, you are going to lose the customer eventually,” Balraj said. “The work started last year and two years back based on the maturity of where retailers are. This year, there is a recognition that they will have to do it at scale.” 

But in order to participate in agentic commerce, retailers first need to ensure that their inventory is surfaceable by the large language models. 

“The existing commerce systems are going to stay in place,” Balraj said, noting that retailers will have to build that middleware for exposing the data to the LLM user interface. 

“Orchestration is the big word here. Somebody has to orchestrate, because agents are not going to work on their own,” Balraj said 

Accurate, real-time product data, such as dimensions, availability and pricing, to name a few, are necessary for agentic commerce to function properly and minimize returns and chargebacks for merchants. 

Protocols are already being created to standardize the way agents interact with outside systems and ingest data. Google has its Unified Commerce Protocol, an open protocol it unveiled at the recent National Retail Federation conference in New York, in tandem with Big Box retailers such as Walmart and Target. Klarna also has its own Agentic Product Protocol, and Anthropic has its Model Context Protocol. 

Product discovery is currently the number one use case that retailers are focused on when it comes to agentic commerce, Michaela Weber, senior vice president of product at Commerce, a publicly traded tech company, told American Banker.  

But to make product discovery a reality, retailers need to make sure their data is ready to be ingested by the large language models that are surfacing their products. To do that, retailers are working with companies like Commerce to deliver their data to AI agents. Commerce owns Feedonomics, a cloud-based product feed management platform. Feedonomics works with Perplexity and PayPal for in-agent purchases on the LLM, and is also working with Stripe and OpenAI to enable in-agent purchases on ChatGPT. 

When it comes to payment, however, brands are still hesitant to turn over the entire shopping experience to agents, because they risk losing the last touchpoint with the customer. 

“AI as a product discoverability tool, 10 out of 10. brands are so excited,” Weber said. “Agentic checkout in the sense that I’m completing a purchase transaction within an AI browsing session in a B-to-C context, it’s a big [to-be-confirmed].” 

That’s not to say all retailers are slow-rolling in-agent payments. Pacsun, a retail and apparel brand that targets teens and young adults, counts agentic commerce as one of its top priorities as it focuses on a unified commerce approach, Pacsun’s Chief Digital and Information Officer Shirley Gao told American Banker. 

“I truly believe that AI is really the future of commerce, Gao said. “The reason that this is the top priority is really because retail has been evolving over the years, along with technology. We moved from the traditional brick and mortar stores, then to online. Then from online, we expanded to mobile apps, and then we started social commerce. This new era of AI… customers are now moving there as well.” 

Pacsun sees opportunities in agentic commerce because LLM operators and protocol developers such as Google and making sure that Pacsun and other retailers are able to own the customer relationship, Gao said. 

“The major difference between social commerce right now versus agentic IA is with TikTok, I don’t have the last mile engagement. The customers are TikTok customers. They check out. I don’t have the first party data. I’m a service provider. We provide the product,” Gao said.

“But with AI, the beauty is really that we own the engagement. We own the relationship,” she said. “From the time you check out, we are the merchant of record. So then it really transformed from AI conversation to a brand customer engagement” 

Pacsun is a customer of Commerce’s Feedonomics, and its products are discoverable and purchasable directly in Perplexity without the customer needing to be redirected out of the LLM to the brand’s website, an important detail that helps ensure a positive customer experience, Gao said. Walmart also has incorporated its checkout and loyalty program into Google’s Gemini, allowing the retailer to maintain key touchpoints with the customer. 

Agentic AI additionally offers retailers opportunities to enhance back-office operations, which is competing for retailers’ investment resources, Kristin Howell, global vice president of retail solution management at SAP, a German enterprise resource management software company, told American Banker. 

“A lot of the retailers we’re working with.. are really focused on use cases that reduce risk, that drive automation, and create a more frictionless experience,” Howell said. 

For example, an agent can take information about a customer order and determine where to best source that inventory from, or more quickly communicate with fulfillment centers to ship orders.  

“AI is a journey; it’s not just one thing,” Howell said. “Like many other technologies – think about search engine optimization, or even ERP deployments. It’s very common to kind of start with things that I would call less risky. Think of the back office, but still very high value, high impact.”Â