EU-native is not a translation layer.

The phrase "EU-friendly" has appeared on every major Amazon PPC vendor's website over the past three years. Pacvue, Perpetua, Teikametrics, Quartile — all four claim European support. All four are headquartered in the United States, were built for the US Amazon marketplace, and were extended to Europe with a localisation pack on top of an architecture that assumes the US is the centre of the catalogue.

Mirox is the inverse. Built in Lisbon, for sellers operating across DE / FR / IT / ES / UK as a primary case rather than an export market. The US marketplace is supported, but it is not the default. That difference shapes most of the architecture.

This is the fourth supporting pillar of Mirox — EU-native — and the one that takes the longest to explain to a buyer who has not yet tried to scale a UK campaign into Germany.

The problem the category does not name

The data on cross-marketplace failure has been public for three years. Ad Badger's international report quantified what every multi-marketplace seller already knew: a UK Amazon campaign copied to the German marketplace without market-specific adaptation loses an average of 42 percent of its click-through rate and gains 28 percent on its ACoS. The figure is similar for UK→FR and UK→IT, more severe for UK→ES, somewhat better for DE→FR.

The cause is not translation. The cause is a stack of decisions that each look small in isolation:

• German buyer search vocabulary uses long compound nouns. The English keyword set translated literally is half as long and twice as generic.
• Italian price elasticity on consumer goods is structurally different from English. A bid that wins the auction in the UK loses on conversion in Italy.
• French marketplace seasonal patterns lag English ones by roughly two weeks during the post-holiday period.
• Spanish placement modifiers have different effective weights because the visual layout of the Spanish product page differs subtly from the English version.
• Days-of-cover for German fulfilment is computed against different FBA fee schedules and different inbound shipment timing windows.

Each of these is a small decision in the campaign architecture. Together they are a 28-percent ACoS hit. A US tool with a language pack catches none of them. A US tool with a thorough localisation pack catches one or two.

What "EU-native" looks like in product terms

Listing what we mean concretely, because the phrase is otherwise marketing puffery.

Per-marketplace pCVR priors

The Bayesian conversion estimator (covered in the intelligence post) runs per marketplace, per category. The German conversion-rate prior is drawn from German data, not from English data with a German label.

This is a small engineering decision with a large effect. A US-first tool typically runs a single prior and applies a per-marketplace adjustment factor. The adjustment factor is wrong in the tail — categories with low data volume in the secondary marketplace get an unstable estimate that propagates into every bid. Mirox refits per marketplace and per category weekly, which keeps the tail stable.

Per-marketplace safety thresholds

The seven layers of Shield carry different thresholds per marketplace. Japan's daily spend anomaly trips at 6× baseline because Japanese seasonality is genuinely more volatile. Brazil and Turkey trip at 3× because their conversion variance is lower and the baseline is more meaningful. The EU marketplaces sit at the default 4×, but the recovery floors and the rolling ROAS windows are tuned per marketplace based on the data we have.

A US tool with a language pack uses US defaults everywhere. The thresholds are set for a marketplace that does not match the seller's marketplace. The system is either too conservative (gates trip during normal operation, the seller disables them) or too loose (gates fail to trip when they should).

Multilingual semantic embeddings, asymmetric

The relevance scoring in Semantic uses an asymmetric multilingual embedding. The query side is computed against the buyer's language; the document side is computed against the product detail page's language. The cross-language pairing — German query against German product page — uses the native-language model, not an English embedding with a translation step in front.

The reason this matters: machine translation drops about 15 to 20 percent of the semantic signal on advertising-domain text, where compound nouns, slang, and product-specific vocabulary dominate. An English-embedding-with-translation approach is consistently worse on relevance than a native-language embedding by roughly that margin. Mirox runs native embeddings for every marketplace language Amazon supports in our deployment.

Currency and rate-limit handling

Every marketplace carries its own currency. Every currency carries its own conversion-rate window. The Amazon Advertising API rate limits are per-marketplace and reset on different schedules. A US-first tool typically treats EU rate limits as a single bucket; Mirox tracks them per marketplace and avoids the rate-limit-exhaustion problem that hits high-velocity EU multi-marketplace operators on the lower-tier vendor stacks.

GDPR-native, not GDPR-retrofitted

Mirox stores tenant data in EU infrastructure by default. The Data Processing Agreement is available on request and was written for Mirox specifically, not adapted from a US template. The aggregate data used for industry-wide modelling is anonymised at the tenant boundary and we do not train cross-tenant models on identifiable seller data.

This is GDPR table stakes for any EU SaaS vendor. We are not branding it as a differentiator. It is mentioned because US vendors occasionally claim GDPR compliance on the basis of a Data Privacy Framework certification, which is a different thing and does not address the data-residency question for buyers who care about it.

Where the differentiation actually lives

The EU-native claim, in product terms, is the sum of the five points above. Each is a small engineering decision. None of them is novel research. The combination, shipped in a single product that defaults to EU behaviour, is rare enough that the only direct competition we have on the EU-native axis is small, recently-founded, and not at scale.

A seller running on DE + FR + IT + ES + UK on a US-first tool has options. They can configure the tool extensively to behave more EU-natively. They can hire an EU-specialised agency to operate the tool on their behalf. They can split their campaigns across two tools and manually reconcile. All three options work to some degree. None of them is the same as a tool that defaults correctly out of the box.

What this does not mean

EU-native does not mean US sellers cannot use Mirox. The architecture supports US bidding and US sellers are welcome. It does mean our case studies, marketing material, and product roadmap prioritise EU multi-marketplace operators — which is a deliberate sales choice based on where our differentiation is largest.

EU-native does not mean we are anti-American. The US Amazon market is the largest and the most competitive. We respect the US-first vendors that built the category. We are simply not trying to compete with them on their home court, and we think the EU is a large enough market that an EU-first specialist deserves to exist.

Where this pillar leads

If you are a US-only seller, the EU-native pillar is the least relevant of the five. The other four pillars — transparency, shadow mode, safety, intelligence, and aligned pricing — still hold.

If you are an EU multi-marketplace operator, this is probably the pillar that closes the sale. The 28 percent ACoS hit is not a marketing number. It is the gap that exists in your campaigns today if you have ever copied a UK campaign to Germany without market-specific adaptation. The Shadow Mode tier will surface the gap on your real account in fourteen days. The founding cohort opens that surface today; the public Shadow tier opens in Q3 2026.

Apply at mirox.pt/beta or join the waitlist at mirox.pt/waitlist.