Mollie, Adyen, Stripe: what embedded payments actually means in European mid-market

The three companies that built the category

European payments processing is dominated, for SaaS and e-commerce mid-market accounts, by three companies that took genuinely different paths to the same neighbourhood. Adyen, the Amsterdam-headquartered processor founded in 2006, built around a global enterprise-grade pipeline that competes for the largest accounts. Stripe, San Francisco by way of Limerick, expanded into Europe through developer-led adoption and SMB self-serve. Mollie, also Dutch and founded in 2004, took the longest and quietest path of the three, building a payments stack for European SMB and mid-market e-commerce that was deliberately easier to wire up than Adyen and locally more compliant than Stripe.

The economics of the European payments market favoured a small number of large players consolidating share, which is what happened. Adyen's 2018 IPO valued the company at €7.1 billion. Stripe's 2023 round valued it at $50 billion. Mollie's 2021 round at $6.5 billion was the structural confirmation that the European-built SMB-and-mid-market positioning was a defensible business in its own right rather than a temporary gap before the global players moved in.

What changed in the conversation from 2023 onwards is that the per-transaction rate, the historical battleground of payment processing, stopped being the most consequential decision for European mid-market buyers. The rates converged inside a narrow band. The real differentiation moved to the layer above: how the processor routes a transaction across networks, how it handles 3D Secure friction, how it reduces declined-payment rates, and how transparent its AI-driven decisions are to the merchant.

What the AI routing layer actually does

A European e-commerce business processing €30M in card payments annually loses roughly two to four percent of attempted transactions to false declines, which is the technical term for legitimate purchases that the issuer's fraud system declined incorrectly. Half a percent of revenue across two to four percent of attempted transactions sits in the false-decline cost. A processor that can reduce false declines by twenty percent through smarter routing is delivering a measurable revenue lift that often exceeds the savings from a marginally lower per-transaction rate.

Adyen's RevenueProtect product, which the company has been iterating since 2019, was the first European-built implementation of this idea at scale. The product looks at the transaction in real time, predicts whether the issuer will decline, and where applicable routes through a network or method that the model believes will produce a higher acceptance rate. The 2024 release of Adyen Uplift extended this to recovering declined payments through targeted retry logic and authentication choices that the merchant did not have to configure.

Stripe's equivalent, branded Stripe Radar for fraud and incorporated into the broader Optimized Checkout Suite for routing, made the same architectural bet with a different commercial framing. Stripe's product is more accessible to smaller merchants because it does not require the integration depth Adyen does. The trade-off is less customisation in the rules layer for merchants who want to override the model's defaults.

Mollie's approach is more conservative architecturally. The company has invested in the methods-and-locality layer first, which matters disproportionately for European e-commerce where iDEAL in the Netherlands, SOFORT in Germany, Bancontact in Belgium, and Klarna across the region account for meaningful shares of consumer payments. The AI routing layer is younger than Adyen's or Stripe's. The compensating strength is that the methods coverage is closer to what a European merchant actually needs.

The AI Act question that surfaces here too

Embedded payments routing systems that make automated decisions about whether a transaction is fraudulent, which payment method to use, or whether to retry a declined transaction sit close to the AI Act's transparency obligations under Article 50 and, for the fraud detection components, potentially within the scope of Annex III high-risk categories depending on how the system is described and used. The merchant is the deployer in most of these arrangements, which means the obligations attach to the merchant rather than to the processor.

The vendors are responding to this in different ways. Adyen publishes detailed documentation on the inputs, model behaviour, and merchant-controllable parameters of its decision systems. Stripe publishes less but offers stronger contractual indemnification. Mollie sits in between. For a European e-commerce business that needs to defend its AI Act posture to its DPO or to an enforcement inquiry, the documentation Adyen provides is the cleanest path. For a smaller merchant who values the contractual indemnification, Stripe's approach is more straightforward.

What the buyer should actually compare

The procurement framework for European embedded payments in 2026 looks different from the framework of 2020. The per-transaction rate matters, but inside a narrow band. The methods coverage matters more for merchants whose customer base is concentrated in countries with strong local payment methods. The routing intelligence matters most for merchants whose false-decline cost is a meaningful share of revenue. The AI Act posture matters for merchants whose compliance function takes its obligations seriously, which is an increasing share of mid-market e-commerce.

A real case from the 2024 to 2025 period: a Dutch e-commerce business at €80M GMV moved from Stripe to Adyen for routing improvements and reported a 1.4 percent revenue lift attributable to the routing changes alone. The Adyen integration cost more, took longer, and required engineering investment Stripe had not. The revenue lift covered the differential inside the first year. The same business would not have made the same move at €15M GMV because the implementation cost would have been larger than the lift.

The category is not converging. It is segmenting. European merchants who know which segment they belong in are making sharper decisions than they were five years ago, and the differences between the three leading processors are now consequential rather than marginal.