Agentic commerce in German-speaking Europe: over-documented by law, invisible to machines

The alphabet soup looks confusing mainly because the players are not all competing with each other. They sit on different layers:

• ▸Discovery — the agent finds your catalogue and understands it: MCP, UCP, AXP
• ▸Checkout — completing a purchase without a redirect into the storefront: ACP, UCP
• ▸Authorisation — is this agent allowed to spend money at all: AP2
• ▸Settlement — how the money actually moves: x402, MPP, cards

The lower three layers are being fought over by Google, Stripe, Coinbase, Visa, Mastercard and Shopware. That is a standards war between corporations and it will be decided without you. Shopware has taken a clear position: UCP, plus its own extension AXP.

Only the top layer is your concern. And not the protocol implementation either, because Shopware now ships that as a free first-party plugin. What matters is what flows through the interface: your product data.

If you sell into Germany, Austria or Switzerland, you carry a documentation burden that merchants in most other markets never think about. GPSR requires manufacturer identification with a postal and electronic address on the product page. The Price Indication Ordinance requires a unit price for anything sold by weight or volume. Textile labelling requires fibre composition. Add the energy label, WEEE and packaging rules on top.

Then the thing happens that happens in nearly every mature Shopware project: all of it ends up as HTML prose in the description field. Sometimes in a shopping-experience block. Sometimes in a free-text custom field somebody created in 2023.

Legally that is often enough, because the information is visible on the product page. Duty discharged. To a machine it is worthless. An agent querying your catalogue over UCP receives a field called description containing a paragraph of prose. It cannot tell marketing copy from fibre composition from a manufacturer address.

GPSR manufacturer data. Company name or registered trade mark, postal address, electronic address, and for non-EU manufacturers the responsible person inside the EU. In practice this is one text block under a heading called product safety. A human reads it. An agent gets a string. Modelled as four to six custom fields bound into the template, the same information becomes centrally maintainable, consistent across channels and machine-readable at no extra effort.

Unit pricing. Shopware has proper fields for this: purchaseUnit, referenceUnit, unitId. In a surprising number of shops they are empty and the unit price lives in the description, or nowhere. For an agent comparing offers, the unit price is the only genuinely comparable figure. It is the sort key. Deliver it unstructured and you are simply not in the comparison.

Fibre composition. Mandatory under EU textile labelling and, in roughly nine cases out of ten, a sentence inside the description. Sixty-five percent cotton as prose is not an attribute. When a shopper asks their agent for a shirt with at least ninety percent cotton, you are not filtered out. You never enter the shortlist.

Dimensions, weight, GTIN. Not a compliance topic, same mechanism. Dimensions and weight stay empty because shipping is billed at a flat rate. The GTIN stays empty because no marketplace channel forces it. Both are fields an agent uses to identify a product and sanity-check it.

Bad product data used to cost you conversion, and conversion is measurable. A visitor arrives, cannot find the information, leaves. You see a bounce rate, a weak detail page, an abandoned cart. There is a signal, and signals can be acted on.

An agent produces no signal. It queries your catalogue, does not find the attributes it needs to answer the question, and leaves you out of the answer. No visit, no abandonment, no row in your analytics. The revenue is simply absent, and nothing anywhere turns red.

There is an amplifying effect on top. Agents do not pick the cheapest offer, they pick the one that can answer the question. Being the only merchant in a competitive set with structured attributes does not win you a few percent. It wins you the mention while everyone else fails to appear at all.

No task force required. An order of operations:

• ▸Measure before you integrate. How many of your articles carry a GTIN? How many have dimensions and a weight? How many have structured mandatory data rather than a text block? In most catalogues at least one of those numbers sits below thirty percent.
• ▸Lift mandatory data out of prose. Move manufacturer data, unit-price fields and material composition into reusable custom fields and bind them into the template. That buys three things at once: consistent legal presentation, clean marketplace feeds, machine readability.
• ▸Normalise variant naming. If the same colour exists in three spellings across two custom fields, no agent can build a reliable offer from it. Neither can your own filter navigation.
• ▸Make stock honest. An agent that buys against a wrong stock level produces a cancellation. A human forgives that occasionally. A system that decides whether to feature you does not forget.
• ▸Only then the protocol layer. Shopware's agentic commerce plugin is free and takes ten minutes. It is the last step, not the first.

I see no value in manufacturing urgency, so here is the other side. Agentic purchases currently account for roughly three percent of online transactions, mostly in the United States and through large marketplaces. Gartner finds that only about eleven percent of US consumers would let an AI make the buying decision. The technology sprints, trust walks. Anyone telling you your revenue collapses next quarter is selling something. Forecasts put roughly one in five digital commerce transactions on AI platforms or agents by 2030, and forecasts for 2030 are forecasts for 2030.

The reason to start now is different: the work pays off whether or not the forecast lands. Structured mandatory data lowers your exposure to competitor warning letters, which in Germany is the real risk rather than regulators. Maintained GTINs and dimensions improve marketplace feeds and shipping calculation. Consistent variants improve your own filtering and conversion. Complete attributes reduce returns.

That is the difference between a bet and an investment. Wiring up a payment protocol is a bet on which protocol prevails. Clean product data pays for itself even if the whole agentic commerce story is judged overhyped two years from now. When only one of two options still earns money if adoption stalls, the decision is not a hard one.