Least-Cost Routing: Sending Payments to the Cheapest Eligible Rail

"Least-cost routing" is one phrase covering three mechanically different things, and most operators researching it are only aware of one. In Australia, LCR is a self-regulatory industry framework for dual-network debit cards. In the United States, the Durbin Amendment's Regulation II mandates something adjacent but distinct — network choice, not cost optimization. Everywhere else, cost-based routing is a commercial decision with no regulatory anchor at all. Building routing architecture on the wrong model — assuming the US works like Australia, or that "least-cost routing" implies a mandate anywhere it doesn't — wastes engineering effort or leaves real savings uncaptured.

This piece separates the three regimes, then covers the ground that applies regardless of which one you're operating under: what data a cost-based router actually needs, when routing to the cheapest rail is a net loss once authorization rate is priced in, the failure modes that show up once you route across rails or acquirers, and the operational cost the routing decision creates downstream.

Three Things "Least-Cost Routing" Can Mean

The practical consequence: an architecture assuming the Australian model — a merchant-set static default a terminal or gateway applies automatically — does not fit the US case, where the issuer's obligation is only to enable the choice and the acquirer or PSP has to build the actual cost comparison. Nor does it fit the generic global case, where there is no enablement mandate and the operator establishes rail eligibility, cost, and authorization data from scratch. The decision math covered later in this piece — route to the cheapest rail only when the authorization-rate delta doesn't erase the saving — is the same in all three. The mechanism for getting there is not.

The Australian LCR Framework

Dual-network debit and how routing is set

Around 85% of Australian consumer debit cards are dual-network debit cards (DNDCs), carrying both the eftpos domestic scheme and a Visa or Mastercard debit brand. On a physical-terminal transaction, the merchant's acquirer and terminal jointly determine which network processes it. Before merchant routing existed, cardholders selected the network by pressing CHQ/SAV (eftpos) or CR (international scheme) at the terminal. Contactless "tap-and-go" removed that manual step — the terminal's chip configuration decides automatically, and it defaulted to the international scheme, a product of how acquirers had configured terminals commercially rather than any technical necessity.

This is not a formal RBA mandate in the sense of a binding rule with a compliance deadline. The RBA endorsed least-cost routing in 2015 and, in November 2017, "strongly supported calls from a range of stakeholders" for acquirers to provide LCR functionality. Parliament recommended regulatory action by April 1, 2018 if banks did not implement it voluntarily — the threat of formal regulation, not a regulation itself, is what drove the industry rollout. Terminal software and hardware upgrades to support merchant-selectable contactless routing ran from late 2018 into 2019. Cardholders retain an override: inserting the card rather than tapping lets them choose the network directly, and merchants using LCR are expected to disclose it.

How the cost difference arises

Interchange and scheme fees for eftpos domestic debit transactions are typically lower than the equivalent Visa or Mastercard debit transaction in Australia. The spread is most pronounced for everyday, low-average-ticket categories: grocery, fuel, pharmacy, quick-service restaurants. For these merchants, card-present debit transactions dominate the volume, and a consistent routing preference to eftpos where cheaper produces meaningful cost savings across the aggregate.

The differential is smaller for card-not-present transactions (e-commerce), which is why CNP LCR was implemented later and required additional work from PSPs — eftpos itself only extended into online/CNP processing in recent years, mirroring the same card-present-first sequencing that shows up independently in the US Reg II timeline below.

PSP implementations

Stripe automatically implements LCR for eligible Australian merchants using dual-network debit cards — no additional configuration required. Adyen supports card-not-present LCR for eftpos in Australia, extending the cost benefit to e-commerce. Most major PSPs operating in Australia now support LCR in some form; merchants with legacy configurations should confirm with their PSP whether LCR is active and whether CNP transactions are included.

The US Debit Routing Regime: Regulation II and the Durbin Amendment

The Durbin Amendment authorized the Federal Reserve to regulate debit interchange and routing. Its implementing rule, Regulation II, contains two separate standards that are frequently conflated: an interchange fee cap, and a network-routing requirement. They apply on different bases.

The routing requirement applies to every debit issuer, regardless of size. Regulation II prohibits issuers and networks from restricting a debit transaction to fewer than two unaffiliated networks, and bars either from inhibiting a merchant's ability to direct routing across whichever networks the issuer has enabled. The $10 billion consolidated-assets threshold that exempts smaller banks applies only to the interchange fee cap — the dual-network routing rule has no size exemption.

Card-present compliance was required by April 1, 2012. Card-not-present was not originally covered. Dual-message networks (Visa, Mastercard) historically dominated CNP processing, while single-message (PIN-debit) networks lacked CNP capability, so issuers routinely enabled only one dual-message network for online transactions — as of 2019, an estimated 25% of regulated issuers still ran single-network CNP debit. The Federal Reserve Board closed this gap with a final rule in October 2022, extending the exclusivity prohibition to card-not-present transactions, effective July 1, 2023.

The mechanical distinction that matters for architecture: Regulation II is a network-choice mandate, not a cost-routing mandate. It forces the issuer to enable two unaffiliated networks and stops anyone from blocking the merchant's ability to choose between them — full stop. It does not obligate any party to actually route to the cheaper network. That optimization is a separate, commercial layer that the acquirer, PSP, or gateway has to build on top of the enabled choice. PSPs serving US merchants market this layer as "intelligent selection" or "managed routing" rather than "least-cost routing" as a formal term — the regulatory vocabulary and the commercial vocabulary have stayed separate even though the outcome for a merchant who configures it is functionally the same as the Australian result: transactions default to the lower-cost available network.

This is the opposite failure mode from Australia. In Australia, the industry built the actual cost-routing capability under regulatory pressure to do so. In the US, the regulation guarantees the choice exists but leaves the cost-routing engineering entirely to the market — a US operator who assumes Reg II compliance means their transactions are already being cost-optimized is wrong; Reg II only guarantees that optimization is possible.

Cost-Based Routing Beyond Regulated Debit Markets

Outside the two regimes above, operators build equivalent routing logic with no mandate behind it at all, typically across three categories.

Domestic scheme routing

Many markets have domestic card schemes with lower interchange economics than international Visa/Mastercard:

• India: RuPay (NPCI) carries zero MDR on debit transactions by government mandate. A merchant processing Indian debit cards routes to RuPay domestically and saves the entire card acceptance cost on that transaction.
• France: Cartes Bancaires is co-badged on over 95% of French cards, with domestic interchange typically lower than international scheme interchange for domestic transactions.
• Belgium: Bancontact dominates domestic debit with similar economics to eftpos in Australia.
• Brazil: Elo and domestic debit rails have different economics from international Visa/Mastercard on Brazilian-issued cards.

The interchange-plus pricing model is essential for this optimization — on blended pricing, the PSP captures any domestic/international interchange differential; on IC+ pricing, that saving flows through to you.

A2A rail routing

Real-time payment rails in many markets carry near-zero or zero MDR: UPI (India) under a zero-MDR mandate for merchant transactions, PIX (Brazil) near-zero for most implementations, PayNow (Singapore) zero-MDR for most merchant setups, and SEPA Instant (EU) typically cheaper than card MDR for A2A credit transfers. The decision rule: if (card MDR − A2A MDR) × volume migrated exceeds the cost of A2A integration plus any conversion-rate decline, add the rail.

Multi-acquirer routing for cost optimization

The broader multi-acquirer routing framework includes cost as one dimension alongside authorization rate, resilience, and geographic coverage. With two acquirers covering the same market, routing logic can direct specific card types or BIN ranges to whichever acquirer has better economics for that transaction profile — the advanced version of cost routing, using ML-based routing to optimize the acquiring path per transaction against a combined cost-and-authorization objective.

Payment Routing Networks: The Rails Available for Cost Optimization

Whichever regime applies, a router only has value if there is more than one eligible rail to choose from. Across the content above, four categories of network recur:

• International card schemes — Visa, Mastercard, Amex. Broad acceptance, generally higher interchange than domestic alternatives.
• Domestic debit/scheme networks — eftpos (Australia), Interac (Canada), RuPay (India), Cartes Bancaires (France), Bancontact (Belgium), and the US regional debit networks (STAR, NYCE, Pulse, Accel) that Regulation II requires issuers to keep enabled alongside the international brand.
• A2A / real-time rails — UPI, PIX, PayNow, SEPA Instant. Not card networks at all; a different acceptance flow with its own UX cost.
• Acquiring paths within the same network — multi-acquirer routing treats two acquirers processing the same card network as separate routable paths, since their negotiated economics and issuer relationships differ even on identical card traffic.

A routing decision is a choice among whichever of these are actually eligible for the specific card, currency, and transaction type in front of you — which is what the next section formalizes.

Routing Decision Inputs

A router missing any one of these inputs will misroute some share of traffic no matter how good its cost model is — most often by sending a transaction to the cheapest rail on paper without checking whether that rail can actually process this specific transaction, which is the first failure mode below.

When Least-Cost Routing Pays vs. When Auth-Rate Optimization Wins

This is where naive cost routing goes wrong. If Rail A costs 1.5% and Rail B costs 1.7%, Rail A looks cheaper. But if Rail B authorizes 2 percentage points more often, the expected revenue per attempt is higher on Rail B for any transaction with positive margin:

Expected net revenue per attempt = (transaction value × margin) × authorization rate − (MDR × transaction value)

Run this per BIN range and acquirer pair using your own settlement and auth data, not vendor-published averages — those figures reflect the vendor's traffic base, not yours. The cheaper-in-MDR option is not always the better economic choice, and the gap between the two only shows up when you segment rather than compare blended totals.

Failure and Edge Cases

Routing to a rail without the right token, currency, or capability. A router built purely from cost and auth history, with no capability gate, will occasionally send a transaction to the objectively cheapest rail that cannot actually process it — no token on file for that acquirer, a currency the rail doesn't settle, or a transaction type (recurring, partial capture) the acquirer doesn't support. These show up as hard technical declines, not the soft, retryable declines the cost model expects, and a naive retry queue that doesn't distinguish the two will burn processing cost retrying something no retry can fix.

Cascading retries. A router that fails over to the next-cheapest rail on any decline — including hard declines — multiplies processing attempts without multiplying approvals, and repeated attempts on a stolen or closed account increase fraud signal exposure rather than reducing it. Decline-code classification has to sit upstream of the routing failover logic, not be bolted on after; see the retry and decline-handling detail in authorization optimization.

Network tokens and Account Updater don't automatically follow a routing change. Network tokens are typically provisioned against a specific merchant ID and acquiring relationship. A cost model that reroutes a stored-credential transaction to a cheaper acquirer may find the token doesn't transfer, forcing a fallback to a raw PAN with worse authorization odds — the opposite of what the reroute was meant to achieve. Account Updater services that refresh expired or reissued card details typically run against one acquiring path; if routing logic moves a recurring transaction to a different acquirer, updater coverage may not carry over, producing "new card" declines the router's own historical data won't explain. The full trade-off between network and PSP tokens for routing flexibility is covered in the network tokens vs. PSP tokens guide above.

Operational Consequences of Multi-Rail Cost Routing

Cost-based routing that spans more than one acquirer or rail creates downstream operational load that the MDR saving has to justify:

• Reconciliation fragments. Each rail or acquirer produces its own settlement file, timing, and fee structure; producing one source of truth across them is consistently underestimated — the underlying build cost is covered in the multi-acquirer routing guide above.
• Chargeback and dispute handling differs by rail. Domestic debit schemes and international card schemes run different dispute timelines and evidence requirements, and A2A rails often carry limited or no chargeback mechanism at all — routing a high-value or high-risk transaction to the cheapest rail purely on MDR, without weighing the dispute-recovery difference, can cost more in unrecoverable loss than it saves in fees.
• Reporting has to segment by rail, not blend. A single effective-cost or auth-rate number across all rails hides exactly the signal that justified building the router in the first place. The metrics to track — per-acquirer auth rate, effective rate by route, routing rule exception rate — are covered in full in the payment routing KPI scorecard linked above.

Scheme Rules and UX Trade-offs

Card schemes have rules governing how branded transactions may be processed, and systematically routing away from an international scheme brand can conflict with acceptance requirements depending on acquirer agreements and scheme licensing terms — review your contracts before configuring aggressive routing preferences.

A checkout with four payment options is not automatically better than one with two well-chosen ones. Offering A2A as a cost-saving alternative to card creates choice, and choice creates abandonment risk for customers who prefer card. High-intent, high-AOV contexts (B2B invoices, subscription renewals) typically see minimal conversion impact; low-intent, impulse contexts can lose more in conversion than they save in MDR. The auth-rate point economics framework applies directly: know the revenue value of your current authorization and conversion rates before optimizing for cost, and evaluate any saving net of the revenue impact it creates.