perspective

The Routing Error

A payment from Nairobi to Shanghai passes through four banks, three compliance perimeters, and loses up to 7% before arrival. The problem isn't speed - it's the route itself.

By Anton Titov, Founder · Plexo Institute

Published 2026-05-17

Cross border payments are expensive not because technology is slow, but because every unnecessary intermediary adds cost, time, and opacity.

Reader Brief

Cross-border payments are expensive not because the technology is slow, but because the routing architecture is wrong. Every unnecessary intermediary in the payment chain adds cost, time, and opacity.

Reading Guide

Four ideas to anchor the read.

The error: a Nairobi-Shanghai payment loses 3-7% across four hops, four compliance perimeters, and four fee schedules.

SWIFT delivers instructions in milliseconds. The delay is what happens at each intermediary: compliance screening, batch processing, time zone misalignment, and sometimes capital-control checks. The median gpi payment settles in under two hours; the slowest routes take more than two days [1].

Three cost layers: remittance fees, B2B infrastructure cost, and trapped working capital.

The routing error is not one cost. It is $59B/year extracted from consumer remittances, 1.6-4% infrastructure cost for B2B operators, and $4T trapped in prefunded accounts to cover settlement delays [3][4][6][17][18].

Why multi-hop persists: it is a trust network, not just a payment rail.

Correspondent banking persists because each hop supplies counterparty trust, legal enforceability, pre-positioned liquidity, and jurisdictional compliance. SWIFT gpi improved tracking and speed, but it did not shorten the route [7][10].

The correction: direct routing, stable settlement assets, and the lesson of Ripple.

Direct stablecoin settlement can replace four sequential screenings with one pre-clearance bundle, one settlement fee, and minutes-long finality. Ripple saw the routing problem but used a volatile bridge asset; the $5.7T adjusted stablecoin transaction market shows that on-chain stablecoin rails already have material scale, while FXC cautions that public data does not cleanly isolate the cross-border payment share [11][12][13][16][20].

The Error

The loss isn't speed. It is the route.

A payment from Nairobi to Shanghai arrives hours or days late, missing 3-7% of its value, and the sender often does not know the final cost until the recipient reports what arrived. SWIFT delivers instructions in milliseconds. The routing architecture itself is the error: each of three or four intermediary banks charges a fee, runs independent compliance checks, and adds processing time.

BIS analysis of SWIFT gpi data shows that payment speed varies by orders of magnitude depending on the route [1]. The fastest routes, usually between major financial centers with direct correspondent relationships, settle in under five minutes. The slowest routes, usually involving lower-income countries with limited correspondent access, take more than two days. The variable is not the messaging protocol. The variable is what each node in the network does before forwarding the payment.

Two-layer routing diagram showing a fast Nairobi to Shanghai message rail above repeated trust work where intermediaries screen, wait, and charge before value reaches the beneficiary. — The routing error is repeated trust work: the instruction can travel quickly while the money rail slows because each intermediary repeats screening, waiting, and charging.

A four-hop payment crosses four compliance perimeters, four fee schedules, and potentially four processing windows.
Hop	Jurisdiction	Compliance layer	Typical fee
Sender -> Correspondent 1	Kenya -> UK	CBK + FCA + UK sanctions	$15-40
Correspondent 1 -> Correspondent 2	UK -> US	FCA + OFAC + Fed wire	$10-25
Correspondent 2 -> Correspondent 3	US -> HK	OFAC + HKMA	$10-25
Correspondent 3 -> Beneficiary	HK -> China	HKMA + PBOC + SAFE	$15-30 + FX margin

The slowest routes expose the hop problem.

BIS finds that prolonged processing times are largely driven by time spent at the beneficiary bank [1]. The bottleneck is not the network. It is what each node does before forwarding the payment: compliance checks, batch processing, operating hours, and capital-control review.

A four-hop payment repeats the same work four times.

Total direct fees in the Nairobi-Shanghai example can reach $50-120 before FX margin. Total compliance screenings: four separate runs of the same payment against overlapping but different sanctions lists. Total processing windows: four, potentially spanning three business days if time zones misalign. A direct stablecoin settlement path requires one compliance check before value moves, one fee, and settlement in minutes.

The Global Cost

The routing error extracts value at three different layers.

The routing error extracts value at every layer of cross-border payments, but the cost looks different depending on who is paying. Consumer remittances, B2B infrastructure, and institutional capital lockup are three distinct cost structures. Conflating them produces misleading analysis.

Layer	Who pays	Visible cost	Market size
Consumer remittance	Individual sender	6.49% average fees + FX margin [3]	$905B in 2024 [4]
B2B infrastructure	Payment operators, fintechs, MTOs	1.6-4% [17]	$32T in 2024 [18]
Cost of capital	Importer, exporter, institutional end-client	3-5 days of locked working capital	$4T trapped globally in prefunded accounts [6]

World Bank Remittance Prices Worldwide covers 367 corridors across 48 sending countries [3].
Receiving region	Average remittance cost	Key driver
Sub-Saharan Africa	7.73%	Fewer correspondents, longer chains, limited competition
East Asia and Pacific	5.65%	Efficient corridors such as China and the Philippines alongside expensive ones
South Asia	4.30%	High-volume India and Pakistan corridors compress costs
Latin America and Caribbean	5.83%	Efficient US-Mexico corridor alongside expensive Caribbean corridors

Layer 1: consumer remittance costs are visible, tracked, and still far from the 3% SDG target.

The World Bank's Remittance Prices Worldwide database reports a 6.49% global average cost across tracked corridors [3]. No region has reached the UN SDG target of 3%. For the $905B in global remittances in 2024, 6.49% translates to roughly $59B extracted annually in fees, FX margins, and intermediary charges [4]. That number is real, but it is only one layer of the routing error's cost.

Layer 2: B2B infrastructure cost is lower per transaction but structurally heavier.

The FSB's G20 Roadmap monitoring reports that the global average B2B cross-border transaction cost is 1.6%, still above the 1% target [17]. In emerging-market corridors, the infrastructure cost paid by payment operators, fintechs, and MTOs is often higher: on/off-ramp fees of 1-2%, compliance infrastructure of 0.3-0.8%, ramp FX margin of 0.5-1.5%, and correspondent or liquidity fees of 0.2-0.5%. On a typical operator margin of ~20%, this 3-4% infrastructure floor leaves only 0.5-1% retained margin. McKinsey data shows why fintech attackers can undercut incumbents while still facing a hard infrastructure floor: incumbents can charge up to 3.4% in P2P, while attackers charge roughly one-fifth of that level [19].

Layer 3: cost of capital is invisible in fee data but dominates institutional trade payments.

For an importer paying through the correspondent banking system, the transaction fee may be the smallest cost component. The real cost is capital lockup: money that cannot be used while it sits in the settlement pipeline. A $1M payment that takes 3-5 business days to settle at a 12-18% annual cost of capital costs roughly $165-250 in implicit financing cost. Across thousands of payments, that becomes a material share of trade value. Globally, an estimated $4T sits idle in prefunded nostro/vostro accounts positioned to cover settlement delays [6].

At institutional scale, the routing error is a capital-efficiency problem.

JP Morgan estimates the B2B cross-border market at $194.6T [5], and the IMF's broader cross-border payment estimate approaches $1 quadrillion [2]. At this scale, the routing error is not primarily a fee problem. It is a capital-efficiency problem.

Why The Architecture Persists

Correspondent banking is a trust network, not just a payment rail.

If multi-hop routing is structurally inefficient, why does it persist? Because correspondent banking is not just a payment rail. It is a trust network. Each correspondent relationship represents verified counterparty trust, regulatory compliance, liquidity, and legal enforceability across jurisdictions. Replacing the routing is easy. Replacing the trust infrastructure is hard.

Feedback loop diagram showing trust, compliance, liquidity, and inertia reinforcing the survival of old correspondent banking routes. — Correspondent banking persists because it is a trust system; a replacement route must prove trust, compliance, liquidity, and switching value together.

Reason	What each hop provides	What a replacement must prove
Trust	Each correspondent is a verified counterparty.	Equivalent counterparty assurance across the full payment path.
Compliance	Each hop satisfies a jurisdictional requirement.	A compliance bundle that supervisors accept before value moves.
Liquidity	Nostro accounts are pre-positioned by corridor.	Reliable real-time settlement and redemption liquidity.
Inertia	Integration cost of switching is high.	Operational simplicity and enough savings to justify migration.

The trust network is shrinking, but no replacement trust layer has emerged at scale.

BIS CPMI data documents a sustained global decline in active correspondent banking relationships [7]. The decline is evident across advanced economies, emerging Asia, Latin America, the Middle East, and Sub-Saharan Africa [8]. In Africa, the decline was sharper: 34.2%, with USD-specific relationships falling 40.9% [6]. As direct correspondent links disappear, remaining routes get longer and more congested. Payments that previously routed through two hops may require three or four.

SWIFT gpi improved speed and tracking, but it did not change the route.

SWIFT gpi delivered measurable speed improvements: 40% of gpi payments credited within 5 minutes, 50% within 30 minutes, 90% within 1 hour, and nearly 100% within 24 hours [10]. But gpi is a transparency and speed layer applied to the existing multi-hop architecture. It does not reduce the number of hops, eliminate intermediary fees, or change routing logic. Tracking a package does not make the delivery route shorter. It makes the inefficiency visible.

The routing error compounds when correspondent relationships disappear.

The Atlantic Council links payment fragmentation partly to increased compliance costs, as banks sever less profitable correspondent links, especially in smaller states with lower volumes [9]. The banks that exit are not replaced. The routes get longer. The costs go up.

What Direct Routing Looks Like

One compliance check instead of four. One fee instead of four. Minutes instead of days.

A direct stablecoin settlement path between two licensed operators eliminates every intermediary in the correspondent chain and can produce better compliance data because full end-to-end counterparty information is assembled upfront.

Comparison diagram showing a traditional correspondent chain with repeated screening, fees, and windows collapsing into one pre-clearance bundle and a direct stablecoin settlement route. — Direct routing collapses sequential trust work into one pre-clearance bundle between accountable origin and destination operators.

Architecture	Route	Compliance pattern	Cost pattern
Traditional correspondent chain	Sender -> correspondent 1 -> correspondent 2 -> beneficiary	Sequential screening at each hop	Multiple intermediary fees plus FX margin
Direct stablecoin settlement	Licensed origin operator -> licensed destination operator	One pre-clearance bundle before value moves	One settlement fee plus local ramp cost

Origin operator Licensed provider clears sender, sanctions, purpose, and source of funds.
Stablecoin transit Value moves directly between licensed operators for settlement.
Destination operator Beneficiary receives local fiat after the compliance bundle clears.

Evidence And Sources

This raw HTML export preserves source visibility for crawler and contractor review. Indexing decision: index, follow.