Agentic commerce explained

09 / 10 / 2026

Your customer is starting to send an AI agent to do the shopping. It searches, compares and buys on their behalf, which changes who you're selling to, and how trust moves through a payment. Here's what agentic commerce means, and what to do now.

6 min.

Artificial intelligence Agentic commerce
Man wearing round glasses using his smartphone

Agentic Commerce in one minute

Table of contents

When the customer isn't the one clicking "buy," how does trust travel through the transaction? That's the question agentic commerce forces and the one that decides who gets chosen next.

Every shift in commerce so far changed where we shop. The high street became the catalogue. The catalogue became the website. The website became the app in your pocket. Each time, the shopper stayed the same: a person, deciding, clicking, paying.

Agentic commerce breaks that pattern. It changes who does the shopping. The person sets the goal; the machine does the rest. That's not a new checkout button it's a new actor in the payment journey.

What is agentic commerce

AI-ready payment experience on a mobile phone

Agentic commerce is commerce where an AI agent carries out part or all of the buying journey for a customer, discovering products, comparing options, and completing a purchase within the customer's instructions.

The scale is already visible. By 2030, AI agents could drive 15–25% of US ecommerce. Today, 58% of consumers already research purchases with AI before they buy. McKinsey estimates transactions flowing through AI agents could reach $3–5 trillion globally by 2030.
Read the white paper

These aren't three versions of the same number. One is a share of a single market (US ecommerce). One is current behaviour (research). One is global transaction value. Together they point the same way: the agent is becoming a buyer, fast.

What changes for merchants

The moment of choice moves.

It used to happen on your site or in your app, the place you control, where you've spent years optimising layout, copy, reviews and checkout. In agentic commerce, the choice can happen inside the agent, before you're even in the room. The agent reads structured data, compares options against the customer's instructions, and shortlists. If your product can't be parsed, it can't be chosen.

So, the job changes. Winning stops being only about SEO, UX and conversion design for humans. It becomes about being three things for machines:

  • Discoverable by AI: your products surface when an agent searches.
  • Understandable by machines: your data is structured, complete and unambiguous, so an agent knows exactly what you're selling and who it's for.
  • Purchasable by agents: an agent can complete the transaction without hitting a wall built for human hands and human eyes.
Woman using a smartphone with her delivery, payment and authentication details confirmed

A worked example. Two merchants sell the same running shoe. One has rich, structured product data: size, use case, materials, stock, return terms, price, all machine-readable. The other has a beautiful page built for people, with the key details locked inside images and marketing copy. When an agent is asked to "find a good trail shoe under €120, in stock, with free returns," only one of them exists. The better page loses to the better data.

woman sitting on the floor using a tablet, with an AI reflection process in progress

 

What changes for banks

Agentic commerce doesn't remove the bank. It makes the bank's role more important.

When an agent initiates a payment, authentication, authorisation, risk scoring and consumer protection don't disappear they get harder and more valuable. Someone has to confirm the customer genuinely stands behind the purchase. Someone has to carry the guarantee that, if it goes wrong, there's recourse.

That someone is the bank. The institutions that build for agent-initiated payments become the trusted layer the whole journey depends on, the part that lets a merchant accept an agent's instruction with confidence and lets a customer delegate without fear.

Who carries the liability in Europe

This is where Europe is different, and where its rules become an advantage.

Under PSD2 and expected to carry into PSD3 a consumer is only liable for a payment they have genuinely authorised, linked to a specific payee and a specific amount. Contract terms alone can't shift that liability onto the customer. A blanket "I agree" buried in an agent's settings doesn't meet the bar.

For agent-initiated payments, that forces something valuable: a tamper-resistant link between the customer's actual approval, the merchant being paid, and the amount being charged. Not "the customer once said yes to this agent." Specifically: the customer approved this purchase, from this merchant, for this amount. That's what "verifiable intent" means, and in Europe it isn't a best practice. It's the law.

Other markets can treat these requirements as friction to engineer around. Built for properly, they become the foundation of trust and the reason agentic commerce can scale in Europe rather than stall on the first dispute.

Man using a smartphone, with a identity and purchase confirmed

Who is accountable across the chain

Agentic commerce works when four roles are clear:

  • The customer sets intent and limits.
  • The agent acts within them.
  • The merchant presents machine-readable products and honours the sale.
  • The bank authenticates, authorises and guarantees recourse.

When those lines are clean, the system is trustworthy. When they're fuzzy, when "the agent did it" becomes an excuse, it breaks. Designing for clear accountability end-to-end is the difference between a demo and a business.

How to get ready in 5 steps

Our white paper sets out a five-step readiness plan. In short:

Read the white paper

1. Make your products machine-readable.

Structured, complete, unambiguous data is the price of entry.
 

2. Decide how customer intent is established.

How does an agent know what the customer actually wants and what they've approved?
 

3. Define agent authority and limits.

What can an agent do on a customer's behalf, and where does it have to stop and ask?

4. Rebuild authentication and fraud controls for a non-human buyer.

The old signals don't read the same way when the shopper is a machine.

5. Plan the payment and recourse path.

How the payment is initiated, authorised, and, if needed, disputed.
 

You don't need all five finished to start. You need to start.

Where this is heading

Three things to watch over the next 12–24 months: the move from human-present to human-not-present agent purchases; the convergence of checkout standards (like the Universal Checkout Protocol) so merchants integrate once, not many times; and Europe's rules increasingly read as a competitive edge rather than a constraint.

Make your move

The shift is here. The real question is who makes it work where the rules are strictest. 

A woman looking at a large screen displaying an AI prompt

Explore agentic commerce

See what's next for your business
A woman reading the Agentic Commerce white paper on her tablet

Read the white paper, free,
no form 

Get your copy now
Innovation centre with demonstration screen

Walk the journey end-to-end at our Innovation Centre in Paris 

Book a visit

FAQ

  • Ecommerce changed where people shop: store to website to app. Agentic commerce changes who does the shopping. The customer sets a goal, and an AI agent searches, compares and buys on their behalf.

  • Two things lined up. AI agents became capable enough to act on a customer's behalf, not just recommend, but search, compare and buy. And consumers started trusting them to do it. What began as AI-assisted research is becoming AI-completed purchasing.

  • Discoverable: your products surface when an agent searches. Understandable: your data is structured and complete, so an agent knows exactly what you sell and who it's for. Purchasable: an agent can complete the transaction without hitting barriers built for human hands and eyes.

  • Because they force trust to be built in. European law requires a verifiable link between a customer's approval, the merchant and the amount. Markets that treat that as friction will stall on the first dispute; building for it is what lets agentic commerce scale.

  • Agentic AI is the broader technology, AI that acts autonomously toward a goal. Agentic commerce is one application of it: AI agents carrying out buying on a customer's behalf.

  • A customer tells an agent, "reorder my usual coffee when I'm running low" and it does. Or "find and book a hotel near this address under €150 with free cancellation." The customer sets the goal; the agent handles discovery, comparison and purchase within set limits. Digital goods and repeat purchases are where it's appearing first, because they're easy to specify and deliver.

  • No. Conversational commerce is a human chatting with a bot to get help buying. Agentic commerce goes further, the agent acts for the customer, completing the purchase on their behalf rather than just assisting. The difference is delegation: one advises, the other acts.

  • The customer remains responsible for purchases they authorised but only those they genuinely approved. In Europe, that means a verifiable link between the customer's approval, the merchant and the amount. If that link isn't there, liability doesn't automatically sit with the customer. For how this works in practice: consent, disputes, chargebacks, see these FAQ.

For implementation detail: SCA, consent, disputes, see the full FAQ on our agentic commerce page

Related resources