Strategy
12 minutes
7.7.2026

Agentic Commerce: Who It Affects – and Who It Doesn't

Which purchases AI agents take over and what shops should do now

Updated on

Benjamin Uhlmann

Ich bin Benni, Gründer der Agentur pan pan. Gemeinsam mit meinem Team unterstütze ich E-Commerce-Unternehmen dabei, ihren Umsatz dort zu steigern, wo ihre Kunden suchen.

Agentic commerce: illustration of AI agents shopping on behalf of people

In brief

Agentic commerce means AI agents that research, compare and buy on behalf of people, within rules the person has set. How much a shop is affected depends not on the industry but on the purchase pattern: rational, recurring purchases (such as printer cartridges or dog food) are taken over by the AI agent, while emotional, one-off purchases (such as perfume or gifts) stay with the human. Current studies show that consumers already trust AI when comparing, but much less when paying. For retailers, clean, machine-readable product data matters most now.

At a glance

  • Agentic commerce hits purchase patterns, not industries.
  • Rational + recurring = agentic zone, where the AI agent buys.
  • Emotional + one-off = the human stays the buyer.
  • Machine-readable product data decides whether a product makes the agent's shortlist.
  • Legally, the AI agent counts as a tool of the user (German law).

Why can agentic commerce transform some online stores while others barely notice it? The answer has little to do with industry or catalog size. It comes down to one question: how much of the human is still in the purchase? The more emotional or financially significant a purchase is, the less often customers hand it over to artificial intelligence. The more interchangeable, rational and recurring it is, the more likely AI agents are to take over the purchasing decisions. This logic fits into a simple matrix. We reviewed current studies from 2025 and 2026 and list the sources below. The article is written for e-commerce and B2B companies that want to know where agentic commerce matters for them and where to start.

What Is Agentic Commerce?

Agentic commerce is the use of autonomous AI agents in e-commerce. They buy on the user's behalf, for consumers or purchasing teams, within rules such as budget, preferences and timing. No more clicking through categories, no more comparing variants, no cart that a person has to fill. The AI agent knows the need and the budget, and it can complete purchases without human intervention in every step. Under the hood, these autonomous agents typically build on generative AI and large language models, and they run entire commerce processes from product discovery to payment.

That is what separates agentic AI from classic chatbots or AI assistants, which mainly answer questions and make recommendations. Conversational commerce, shopping through dialogue, is only the stage before: the human leads the conversation and triggers the purchase. In traditional commerce, a person also clicks through every step. In agentic commerce the software can act independently, as long as it stays within the rules it was given. Consumers state their wishes in natural language, for example “reorder the dog food I always buy, but not above 40 euros”, and the agent uses past purchases and preferences to turn that into concrete tasks.

The economic dimension is significant. McKinsey estimates that by 2030, between 3 and 5 trillion US dollars of goods sales worldwide could be orchestrated by AI agents, not counting services and B2B marketplaces.¹¹ That is a forecast, not a measurement, but it shows why retailers take agentic commerce seriously.

Agentic Commerce vs. Traditional E-Commerce

Traditional e-commerce is built for human visitors: category pages, filters, product pages and a cart. Agentic commerce is built for agent interactions. A shopping agent does not browse, it queries data. Where online shopping for a person means opening one tab after another, a single agent can compare multiple retailers in seconds and report back with a shortlist or a finished order.

For retailers this changes what counts. In agentic commerce, structured product data and clean interfaces matter more than the layout of a product page, and agentic commerce tools such as protocols, feeds and APIs matter more than another banner. Your customers are no longer just humans. Agentic commerce is therefore not e-commerce with a chatbot on top, it is digital commerce with a new kind of customer, and as agentic commerce continues to develop, more platforms will add native checkout.

How Agentic Commerce Works

For many online shops this still sounds like the future. It is not, because the technical foundation already exists. In September 2025 OpenAI and Stripe introduced Instant Checkout in ChatGPT and the Agentic Commerce Protocol (ACP), initially for users in the US.⁶ In January 2026 Google and Shopify followed with the Universal Commerce Protocol (UCP), an open standard that lets AI agents and merchants agree on capabilities and execute transactions. It was developed together with retailers such as Etsy, Wayfair, Target and Walmart.⁵

The consultancy Bain distinguishes three types of AI agents: third-party shopping agents such as ChatGPT, Perplexity or Gemini, which browse merchant sites and compare offers; agents inside a retailer's own environment, such as Amazon's Rufus; and agents that shop across multiple retailers, such as Amazon's “Buy for Me”.¹ Which type starts the purchase determines who keeps the customer relationship in the end.

In practice the buying journey changes first: research, comparison and selection move into the dialogue with the AI before the customer ever sees a retailer's site.¹¹ The agents pull product data, prices and availability, match them against the user's wishes and finally trigger the order and the payment. They can monitor inventory levels and prices in real time, which only works when the shop delivers reliable real-time data. Technically this runs through APIs, which let agents connect to existing systems such as the shop, the inventory system and payment systems. The more of these steps run automatically, the more competition shifts from the storefront to data quality.

Benefits of Agentic Commerce for Companies and Retail

For retail companies, agentic commerce is first of all a matter of operational efficiency. AI agents are not tied to opening hours and work around the clock, they automate routine tasks and workflows, and they relieve teams in purchasing, customer service and marketing. Subscription management and replenishment are natural fits. New products can also reach the market faster when product data and offers are pushed automatically to channels and agents. On the customer side, agents can provide personalized support along the entire customer journey because they know purchase history, preferences and budget. That makes AI shopping more individual, but it can also change the relationship between consumers and companies: when an agent stands between customer and shop, it helps decide which brands become visible at all.

For retailers this means: anyone who used to compete for human attention through design, copy and campaigns now also competes for selection by software, where decision making is based on price, availability, delivery time and the quality of the stored data. That is no argument against marketing, but it shifts the weighting, especially for low-risk purchases the customer simply wants to obtain as easily as possible.

The Protocols Behind Agentic Commerce

For AI agents and shops to talk to each other, they need shared standards. Three names are worth knowing:

ProtocolBy whomWhat for
ACP (Agentic Commerce Protocol)OpenAI and StripeBuying directly in chat (Instant Checkout in ChatGPT); merchants provide a product catalog
UCP (Universal Commerce Protocol)Google and ShopifyOpen standard for product discovery, cart and checkout between AI agents and merchants
AP2 (Agent Payments Protocol)GooglePayments: digital mandates prove cryptographically that a person authorized the AI agent to buy

The difference matters for merchants: with ACP you pass your product data to the platform and let the checkout happen in the chat.⁶ The Agent Payments Protocol is about making sure that every payment by an AI agent rests on a verifiable instruction from the user.¹⁰ The standards partly complement each other but also compete over who controls the interface to the customer. For merchants this means a clean data foundation pays off across all protocols, and choosing one protocol is less urgent.

The Matrix: Two Axes, One Simple Logic

The matrix shows where agentic commerce takes hold and where it does not. It has two dimensions. The first axis asks whether a purchase is made rationally or emotionally. The second asks whether it is a one-off or recurring purchase. That gives four fields:

Recurring purchaseOne-off purchase
RationalAgentic zone · the AI agent takes over
Printer cartridges, charging cables, dog food, drink crates, workwear
Gray zone I · research purchase
Printer, laptop, WPC decking
EmotionalGray zone II · preference purchase
Wine, lipstick, partly perfume
Human zone · the human stays the buyer
Sneakers, perfume, sweater, gifts, sofa

Agentic Zone vs. Human Zone

CriterionAgentic zoneHuman zone
Purchase patternrational, recurringemotional, one-off
ExamplesPrinter cartridges, dog foodPerfume, gifts, sneakers
Role of the brandlow, interchangeablehigh, decisive
RecommendationSubscriptions & agent interfacesStrengthen the brand experience

Where the AI Agent Takes Over

Printer cartridges, charging cables, dog food, drink crates and workwear are the textbook examples of the agentic zone, and agentic commerce will reach them first. People do not buy these anew, they reorder. The brand is usually set by the first device or the first contract, not by a conscious preference. There is no research moment, no comparison, no joy in browsing. Anyone who spends time here usually experiences it as a burden, which makes these routine transactions the obvious candidates for an AI agent.

For retailers whose core range sits in this area the consequence is direct: once an AI agent has access to purchase history and preferences, there is little reason for a person to go through the ordering process manually. The touchpoint with the brand disappears, not abruptly but gradually. The most sensible answer is therefore not to work against the AI, but to make reordering and subscriptions so easy that the channel stays with the retailer.

Where the Human Stays the Buyer

At the other end are sneakers, perfume, a sweater, a gift, a sofa. Here the decision itself is the value. People compare styles, test scents, think about what fits their personality. An AI agent can filter options, but it cannot feel which perfume suits someone's presence or which gift will really delight. For shops in these categories agentic commerce means one thing above all: the traffic stays human as long as the brand itself is reason enough to visit. Zalando and Flaconi are good examples, because their business model lives on exactly the indecision that an AI agent cannot resolve.

Agentic commerce does not hit industries. It hits purchase patterns.
Benjamin Uhlmann, founder of pan pan

The Gray Zone That Is Often Overlooked

It gets interesting where the matrix is not clear-cut. Printers, laptops and WPC decking are rational but one-off. People buy them rarely, but with real research effort. As long as that deliberate comparison moment exists, the human stays at the wheel. This is exactly where artificial intelligence is already strong today, as a decision aid that helps people compare options. A study by Frankfurt School and the University of Hamburg, published in Marketing Science, shows that for complex purchases with high information needs ChatGPT can already be a relevant channel, while for simple standard products it trails most other channels.⁴ The study also puts the overall share into perspective: across 973 online shops examined, traffic from large language models made up less than 0.2 percent.⁴

The reverse holds for wine, lipstick and partly perfume: these purchases repeat but are emotionally charged. Recurring does not automatically mean at risk. It depends on whether the repetition comes from convenience or from genuine preference. This nuance is what makes the matrix more useful than a split by industry.

What Consumers Actually Allow AI Agents to Do Today

There is a gap between advising and paying. According to the KPMG Consumer Snapshot 03/2026, conducted by the EHI Retail Institute, 73 percent can imagine using an AI agent to compare products, prices and offers. When it comes to a binding step such as payment, the figure drops to 45 percent.² McKinsey sees the same pattern in Europe: in a survey across Germany, France and the UK, 84 percent use AI in daily life and 38 percent for product research or purchase decisions, and usage falls the closer people get to checkout.³ For context, the McKinsey sample covers 749 respondents.³

People want to keep control. 79 percent want to be able to step in at any time, and 72 percent consider it essential to set rules, budgets and exclusion criteria.² On trust, the second look matters: 64 percent consider general AI models more objective and less influenced by commercial interests, and only 7 percent prefer the AI of a retailer or manufacturer.² At the same time, according to Bain, consumers currently trust retailer-owned agents three times more when it comes to completing transactions.¹ Our reading: for selection, neutrality counts; for paying, the familiar retailer counts. For shops this means the decision often falls before the storefront, while the purchase still often happens there. That also affects conversion: if you do not show up in the agent's selection, you never get the chance to close the sale.

Machine-Readable Product Data: The New Ticket In

An AI agent evaluates products by data, not by ad copy. McKinsey names three prerequisites for merchants: product catalogs must be readable for agents, they need semantic metadata so the AI understands the customer's intent, and merchants should offer robust interfaces (APIs) through which agents interact with their systems.¹¹ In practice this means structured product data with complete attributes, clean variants, correct prices and real-time availability. The structured data should be machine-readable, ideally from one central source that feeds the shop, the feed and the interfaces alike. Product data quality therefore decides directly whether a product makes it into the AI agent's selection at all, and the same applies to the business data behind prices, stock and delivery terms.

Governance comes on top: when software triggers payments, there have to be clear rules on who may decide what. McKinsey recommends, among other things, tiered permissions, traceable decision logs and the ability to reverse purchases, and considers auditable logs possibly mandatory in the future.¹¹ For retailers and payment providers this is a sign that documenting orders placed by AI agents is likely to become standard. Payment networks and payment providers will also need authentication systems and fraud detection that can tell trusted agents from malicious bots. Data protection and compliance matter too: as soon as agents work with customer data, purchase histories and payment information, the usual privacy requirements apply, and companies should clarify early which data they pass on to third parties and their software.

B2B Commerce as a Pioneer

In B2B commerce the logic of the agentic zone applies with particular force, because purchasing follows budgets, rules and recurring demand. A study by ECC KÖLN and Arvato Systems among 100 executives and employees from retail, wholesale and industry, surveyed in May 2026, shows the direction: 86 percent are convinced that autonomous AI agents will permanently change existing business models. 63 percent expect price, availability and sustainability to outweigh brand image, and 74 percent expect AI platforms to gain relevance.⁹ The sample is small and heavily weighted toward manufacturers (76 percent), so the results are a mood snapshot rather than a market measurement. For B2B companies this means structured, complete product data becomes more important than the brand presence, and anyone who stores terms, tiered prices and delivery times cleanly is easier for purchasing agents to find.

Who Is Liable When the AI Agent Buys?

This section reflects German law, which is where we operate. Legally there is no special status for AI. According to the German Händlerbund, an AI agent is not a representative within the meaning of sections 164 ff. of the German Civil Code (BGB) but a tool of the user, and its actions are attributed to the user. In case of a technical transmission error, a challenge is usually possible. If the AI agent makes a poor decision on the merits within its instructions, the user is generally bound. If it acts outside its instructions, for example beyond a price limit, it depends on the individual case.⁷ The right of withdrawal remains for consumers, and it does not apply to purchases by businesses.⁷

A new obligation for all shops comes on top: since June 19, 2026, merchants who conclude contracts with consumers in Germany must offer a clearly visible and permanently accessible withdrawal button (section 356a BGB, implementing EU Directive 2023/2673).⁸ For orders placed by AI agents, the Händlerbund recommends clear product information, clean ordering processes and good documentation, because from the outside it is often not visible how the AI was configured.⁷ This is an assessment, not legal advice.

What This Means for Your Range

The strategic question is not whether agentic commerce affects you, but which part of your range sits in which quadrant. An electronics retailer sells laptops that are researched and printer cartridges that are simply reordered. Both in the same shop, with completely different exposure to agentic commerce. Going through a range, you usually find three types:

  1. Pure reorders: Think about subscription models or your own ordering interfaces for AI agents instead of losing the channel.
  2. Brand and discovery products: Keep strengthening the emotional side of your brand presence.
  3. Gray-zone products: Watch which way buying behavior moves, and structure product information so that an AI finds the right facts when it compares.

In practice we start with three steps for clients. First, work out revenue per quadrant, so it is clear how large the agentic zone in your own shop really is. Second, keep product data in Google Merchant Center clean. Google is aligning the rollout of UCP with AI Mode and Gemini, so a complete and error-free product catalog is, in our view, the most pragmatic preparation.⁵ How to fix disapproved products is covered in our guide Fixing Disapproved Products in Google Merchant Center, and with a product feed audit we check the data systematically. Third, track traffic from AI sources separately to see whether and when relevant volumes arrive. The right agentic commerce tools are secondary here; what matters is that data, systems and responsibilities are settled inside the company.

Agentic Commerce FAQs

What is agentic commerce in simple terms?

Agentic commerce means an AI agent researches, compares and buys on behalf of a person, on its own, within rules the person has set.

What is the difference between AI assistants and AI agents?

AI assistants answer questions and make recommendations, and the human decides. AI agents carry out tasks on their own, up to the order and the payment.

Do AI agents already buy on their own today?

In part. Instant Checkout in ChatGPT launched for users in the US.⁶ In Europe, according to McKinsey, the phase before the purchase dominates so far: the closer people get to checkout, the less AI they use.³

Is agentic commerce secure?

Security rests on clear permissions, authentication and verifiable mandates, which is the idea behind the Agent Payments Protocol.¹⁰ Companies should still define rules for what an agent may buy and keep logs of its orders.¹¹

Which shops does agentic commerce hit first?

Not particular industries, but ranges with rational, recurring purchases such as consumables, pet food or reorders.

What product data do AI agents need?

Complete, structured and current product data: attributes, variants, prices and availability, ideally in real time and retrievable through interfaces.¹¹

Is a purchase by an AI agent legally valid?

Generally yes under German law. The AI agent's action is attributed to the user, and the right of withdrawal remains for consumers.⁷

Conclusion

Agentic commerce does not hit industries, it hits purchase patterns. Where a purchase is a pure commodity and no emotion is involved, AI becomes the natural default. Where there is something to discover or a brand carries real meaning, the human stays the buyer, at least as long as brands actively maintain that meaning. For retail companies, one thing holds in both cases: clean, machine-readable product data is the foundation for staying visible to people and to AI agents. If you have not yet sorted your range into the matrix, start now. Not out of worry, but out of clarity about where action is needed.

Quellen

Latest Articles

Knowledge Base

Interviews, tips, guides, news and best practices on the topics of search engine optimisation, SEA and digital strategy.
SEA
14 minutes

How to Structure Performance Max for Online Shops

How many Performance Max campaigns a shop needs, how we sort products into top sellers, challengers and slow sellers with a script, and why target ROAS and budgets work differently since August 2026.
To the article
Strategy
10 minutes

Contribution Margin 3 in E-Commerce

How to calculate contribution margin 3 per order, why new and existing customers need separate numbers, and how marginal CM3 steers your ad budget.
To the article
7 budget killers in SEO projects and how to avoid them
SEO
8 minutes

7 Budget Killers in SEO Projects – And How to Avoid Them

From audits no one implements to tool overkill and perfectionism in the wrong places: The 7 most common budget killers in SEO projects – and how to protect your budget.
To the article

Get in touch

We'll get back to you with an answer within 24 hours — we promise.

By entering your data, you agree to the Privacy Policy I agree with Pan Pan.

Great, we have received your message!

We are looking forward to working with you to achieve your success!
Oops, something went wrong!

Please try again or contact us directly via email.