Cross-border payments: the architecture choices that compound
The decisions that quietly determine whether a global payments stack throws off margin or burns it. Multi-acquirer routing, FX strategy, and the case for treating cross-border as architecture, not procurement.
Cross-border payments are usually treated as a procurement problem: pick a processor that supports international cards, accept the cross-border fees, move on. That framing leaves seven figures on the table for any platform processing meaningful international volume. The right framing is architectural — and the architectural choices compound.
What “cross-border” actually means
A transaction is “cross-border” when the issuer’s country differs from the acquirer’s country. Not when the customer is foreign. Not when the card has a different billing address. The classification is determined by the bank that issued the card and the bank that accepted the transaction.
This is the lever. A US-based platform with a US acquirer treats every Visa transaction issued by a non-US bank as cross-border — paying cross-border interchange (typically 1.0%–1.6% on top of standard) and international scheme assessments. The same transaction routed through a European acquirer to a European-issued card is domestic. Same customer, same product, same payment method. Different basis points.
The three architectural choices that matter
1. Single global vs multi-acquirer
The default architecture is single-acquirer: one processor relationship handles every transaction regardless of geography. It is simple. It is operationally light. It is also, at scale, expensive.
Multi-acquirer architecture maintains a regional acquirer in each major market: US for US-issued cards, EU for EU-issued cards, UK for UK-issued cards, APAC for APAC-issued cards. A routing layer at the gateway determines which acquirer handles each transaction.
The economics: cross-border interchange disappears for transactions routed regionally. Approval rates rise (regional issuers are more comfortable with regional acquirers — there is less fraud signal in a domestic transaction). Negotiating leverage compounds because no single acquirer has a monopoly on your volume.
The trade-off: implementation complexity, multiple vendor relationships, more sophisticated reconciliation. Worth it above $50M annually with meaningful international exposure. Not worth it below.
2. Centralized vs distributed FX
Currency conversion is the second-largest line item in cross-border, and frequently the largest. Default behavior is processor-managed FX: the processor converts at their published rate (typically 1%–3% above wholesale) and passes the cost through.
Alternative architectures:
Multi-currency settlement. Hold balances in major settlement currencies (USD, EUR, GBP, sometimes others). Convert on your schedule, with your treasury, at rates closer to wholesale. Suited for platforms with predictable currency flows above $10M annually per currency.
In-house FX with a bank counterparty. For platforms with non-trivial currency exposure, negotiating an FX line with a bank or non-bank counterparty (Convera, Wise Business, Corpay) can save 100–250 basis points on FX margin. The operational overhead is real; the savings are larger.
Hybrid. Most reasonable architectures use processor-managed FX for low-volume currencies and a more sophisticated structure for the top three or four currencies.
For most cross-border platforms, FX margin is a bigger line item than card processing optimization. Sizing both early lets you attack the larger lever first.
3. Issuer-routing intelligence
Beyond geographic routing, there is issuer-level routing intelligence. Some issuers have higher approval rates with certain acquirers. Some accept network tokens; some don’t. Some respond to particular 3DS configurations more favorably than others.
Sophisticated platforms collect this data and route accordingly. The infrastructure is non-trivial — it requires capturing approval/decline data, building a model of issuer-acquirer-feature interactions, and feeding the model into routing decisions in real-time. But the approval-rate gains compound: a 1.5% improvement in approval rate on $200M in annual volume is $3M in recovered revenue.
A simplified case
A SaaS marketplace processes $100M annually with the following geographic split: 60% US, 25% EU, 10% UK, 5% rest of world. Currently routes everything through a single US acquirer.
Cross-border interchange leakage: 40% of volume × ~120 bps cross-border premium = ~48 bps weighted on total volume. On $100M, that’s $480,000 per year.
FX margin leakage: 35% of volume in non-USD ≈ $35M, at 2% processor FX margin = $700,000 per year. Negotiable down to ~0.5% with a moderately sophisticated structure.
Combined recoverable: roughly $900,000 per year, recurring. The implementation cost is meaningful — 4–6 months of work, regional acquirer onboarding, gateway integration, reconciliation rebuild — but the payback is comfortably under a year. After payback, the savings compound.
The architecture-vs-procurement framing
Treated as procurement, cross-border looks like a fee discussion: “what rate do we get?” Treated as architecture, it looks like a structural decision: “where does volume route, and what happens to the unit economics over time?”
The architectural framing pays back over years. The procurement framing pays back at the negotiation table once and stops.
When this is wrong
Below $20M in annual cross-border volume, the architectural complexity rarely pays back. Single-acquirer with disciplined fee negotiation is the right answer.
Between $20M and $50M, the answer depends on the geography concentration. High concentration in one foreign region can justify a single regional acquirer addition. Wide distribution rarely does.
Above $50M, architecture compounds. Above $100M, it is malpractice not to engage with it.
Cross-border architecture is engagement-by-engagement work. The framing here is general; the right answer depends on your actual volume, geography, and currency mix. For a tailored analysis, schedule a strategy call.
