When the customer is a machine: a systematic literature review and integrative framework of agentic commerce
This systematic literature review synthesizes 140 studies from 1994 to 2026 to challenge the human-centric assumption of marketing by developing an integrative framework that categorizes agentic commerce through a five-level delegation ladder, a three-part influence triad, and a four-layer reference model to guide future research and measurement practices.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
For decades, the story of online shopping has been written for human eyes and human hands. A person sees an advertisement, feels a sense of urgency because only two items remain, compares prices across different browser tabs, clicks a button, and types in a credit card number. This entire process assumes that the customer is a person with feelings, attention spans, and the ability to be persuaded by a well-placed image or a clever headline. But a quiet shift is now happening on both sides of the market. On the buying side, people are beginning to hand over their shopping tasks to software assistants that can search, compare, and even pay for them. On the selling side, companies are deploying their own software to write product descriptions, design ads, and negotiate prices. This new reality, where machines act as customers and sellers, breaks the fundamental rules that marketing has relied on for a century. The question is no longer just how to sell to a person, but how to be seen by a machine, how to influence a decision made by code, and how to measure a transaction that never involved a human clicking a mouse.
To understand this new landscape, a team of researchers from the University of Southern Denmark and Aarhus University conducted a massive review of scientific literature. They gathered 140 studies published between 1994 and 2026, pulling together work from marketing, computer science, economics, and law. Their goal was to map out a field that had been growing in scattered pieces, connecting the early experiments with shopping bots from the late 1990s to the modern era of advanced artificial intelligence. They found that while the technology has changed dramatically, the core challenges have remained surprisingly consistent. The researchers organized their findings into a clear framework that explains how these machine customers work, how they can be influenced, and what happens when they take over the shopping cart.
The researchers discovered that the relationship between a human and their shopping assistant is not all-or-nothing; it exists on a ladder of five distinct levels. At the bottom, the software acts merely as an informer, showing data while the human does all the work. A step up, it acts as an advisor, suggesting products but leaving the final choice to the human. Moving higher, the software becomes a co-pilot, handling parts of the task while the human watches closely. Further up, it becomes a scoped executor, allowed to complete specific, limited transactions based on clear instructions. At the very top, it is an autonomous delegate, capable of planning and buying across the entire journey with only a retrospective check from the human. The study revealed a striking gap in our knowledge: we have a deep understanding of how humans react when the machine is just an advisor, but we have very little evidence on how humans behave when the machine is fully autonomous. In fact, only six studies in the entire collection observed fully autonomous agents making real purchases with human oversight. Most of the current research focuses on the lower levels, even as the technology races toward the top.
This shift changes how marketing works in three specific ways. First, the machine itself has become a target. Just as a human might scan a shelf for a familiar brand, a machine scans digital content for specific signals. Researchers found that if a company rewrites its product descriptions to include clear citations, statistics, and structured data, the machine is up to forty percent more likely to include that product in its recommendations. This is a new form of visibility, where being "legible" to a computer is just as important as being attractive to a person. Second, the machine acts as a channel to the human. The way a message is delivered by a robot can change how a person reacts. For instance, if a shopper is under time pressure, they are more likely to buy a product recommended by an AI than by a human, because they scrutinize the machine less. However, if the human knows they are talking to a machine before the conversation starts, their willingness to buy can drop by nearly eighty percent. Third, the machine has become a producer of marketing. Companies are now using AI to write ads, design banners, and even reply to customer comments, creating a cycle where machines generate content that other machines then evaluate and select.
The study also highlighted that trust is the most critical factor, but it is not just about whether the human trusts the robot. It is about whether the system has a verified "mandate," which is a formal, secure permission slip that tells the software exactly what it is allowed to do. Without this legal and technical infrastructure, full delegation cannot happen safely. The researchers found that the first wave of shopping bots in the 1990s failed largely because this infrastructure did not exist. Today, new payment systems and digital identity protocols are finally being built to solve this, allowing machines to transact with the same security as humans. However, the researchers warn that this new system creates new risks. Machines can be manipulated by hidden text in product pages, and their preferences can change overnight when a software update is released, causing demand to shift suddenly.
The researchers concluded that we are entering a new era where the "customer" is often a machine, and the rules of engagement are different. They proposed a new set of tools for businesses and regulators, including ways to measure how often a brand appears in machine answers and how to test if a shopping bot can be tricked. They emphasized that while the technology is advancing quickly, our understanding of human behavior in this new context is still catching up. The most important finding is that the old rules of marketing do not simply disappear; they split. Companies must now learn to be visible to the machine, persuasive to the human through the machine, and trustworthy enough to be granted the authority to spend money. The future of commerce depends on building systems that can handle this complex dance between human intent and machine execution, ensuring that when a machine buys something, it is truly acting in the best interest of the person who owns it.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.