B2A2C, or Business to Agent to Customer, describes a model of commerce in which a business interacts with its customer through an artificial intelligence agent. The assistant helps the customer explain what they need, discover and compare products, make a purchase, and track their order by interacting with the retailer’s systems.
The bookseller remembered you
He remembered the novel he had recommended a few weeks earlier. Your enthusiasm, or your disappointment. He knew you liked historical fiction, but he also knew that the book you bought in December was for your father.
Your bookseller didn’t just know his books. He knew his readers.
Sometimes he got it wrong. You would come back and tell him that a book had left you cold. He would ask a few questions, get a better sense of your tastes, then lead you towards another shelf. The conversation would pick up where it had left off.
The internet has given us access to millions of titles. But it has often deprived us of that relationship. We replaced advice with filters, conversations with search engines, and the retailer’s memory with browsing histories.
Yet artificial intelligence may bring us back to that older form of commerce. With one difference.
This time, the customer does not necessarily walk through the door. They ask ChatGPT, Claude or Gemini. They explain what they are looking for, say who they are buying for and set a budget. The assistant asks questions, compares and suggests. And, when the user allows it, it can retain some of their preferences so the conversation can continue.
It is a kind of digital personal shopper, available at any hour, to which we sometimes find ourselves paying almost human attention.
Conversation is returning to the heart of commerce. Only the person answering us is no longer necessarily the person selling to us.
What if your best product was never recommended?
Imagine that our bookseller has just discovered a remarkable novel from a small publishing house. He knows which readers to recommend it to. He knows its strengths, its distinctive qualities and even its flaws. He could talk about it for hours.
But a potential customer asks their assistant to find a historical novel for their father. The AI understands the request. It checks the information available to it and selects a few books.
The bookseller’s novel might have been the perfect choice. Yet it does not appear. Not because it is less good. Nor because it costs too much.
Simply because the assistant cannot access the information it needs to discover the book, understand why it might appeal to the reader or buy it from that retailer.
The reader is satisfied. They have found a book. The bookseller will never know he could have sold it to them.
The risk is no longer just that customers leave your shop. It is that they find what they need without ever knowing you existed.
A revolution we can already measure
It would be easy to think this is still a distant prospect. Yet habits are beginning to change.
According to an international study published by the National Retail Federation and IBM in January 2026, 41% of consumers surveyed said they use AI assistants to research products.
According to figures published by Adobe for July 2026, visits to US retail websites from AI sources had a conversion rate 60% higher than that of the other channels studied.
This does not mean fully autonomous purchasing has become widespread. But assistants are already involved in product discovery, comparison and the preparation of purchasing decisions.
A study published by Boston Consulting Group on 21 October 2026 found that 68% of retailers surveyed expect agentic commerce to have a significant or transformational impact over the next two years. Yet only 18% consider their marketplace ready to interact with AI agents.
There remains a considerable gap between recognising the change and being able to respond to it.
Meanwhile, OpenAI is developing product discovery and commercial interactions within ChatGPT. Google is building its own mechanisms to connect search, advice and purchasing. New assistants can browse the internet and carry out certain actions on behalf of their users.
The question is becoming urgent: when a customer entrusts their next search to an assistant, will your products be among the possible answers?
How many languages must you speak to sell a book?
Imagine for a moment that our bookseller had to change language depending on who entered the shop. With each person, he would need to learn new expressions, understand different rules and adapt to different practices. And when the conversation led to a purchase, even the formalities would change.
How much time would he have left to read, discover new authors and advise his customers?
This little story resembles the technical challenge facing retailers. The major AI players are developing different protocols: some structure commercial exchanges (ACP, UCP), others provide access to services and interfaces (MCP, WebMCP). Payments are also seeing new approaches (AP2, MPP).
These initiatives do not all serve the same purpose. Some may complement one another. But each has its own conventions, integration requirements and pace of development.
A good-looking digital catalogue is no longer enough. Assistants also need to be able to access reliable information, check stock, understand product variants, build a basket, support payment and retrieve order details.
The simpler commerce becomes for the consumer, the more complex it risks becoming for the seller.
Cartmelai: making the technology invisible, not the retailer
This is precisely the contradiction Cartmelai aims to resolve.
The French company is developing infrastructure that enables retailers to make their sales systems accessible to AI assistants without having to build and maintain a separate integration for every new environment.
It acts as an abstraction layer between e-commerce platforms and agentic commerce protocols. But its ambition goes beyond making a catalogue discoverable.
Cartmelai aims to support the entire commercial journey: from product discovery and recommendation to building the basket, payment, ordering and tracking.
Let us return to our reader. They ask their assistant for help choosing a gift. The assistant can check the bookseller’s available offers, confirm that a book is in stock, give its price and help complete the purchase.
A few days later, the reader returns to the same conversation: “By the way, where is the book I ordered?” Thanks to the connection with the retailer’s systems, the assistant can access the authorised order information and provide an up-to-date status.
The relationship does not end at payment. It continues after the purchase.
Making these experiences possible means connecting technologies that were not all designed to work together. That is precisely the role Cartmelai seeks to take on.
