Payment Orchestration: The Switchboard of Modern Commerce

Article # 57: How payment orchestration became the most consequential layer in fintech — and why AI agents just raised the stakes.

Apr 09, 2026

TLDR:

Payment orchestration — the infrastructure layer that routes, authorizes, and optimizes every transaction across multiple PSPs, rails, and methods — is growing at 18–26% CAGR toward $2–10B by the early 2030s.
The average enterprise merchant now integrates 6–8 payment methods across multiple providers. 81% of global enterprises run omnichannel strategies requiring 8+ methods per platform. Manual routing can’t scale.
AI-powered routing engines are cutting authorization declines by up to 12% in high-volume verticals. Real-time fraud detection, deployed by 58% of orchestration platforms, has reduced transaction failure rates by 29%.
On March 18, 2026, Stripe and Tempo launched the Machine Payments Protocol (MPP) — the first open standard for AI agents to autonomously request, authorize, and settle payments. Visa extended MPP to card-based payments the same day. 100+ services were in the payments directory at launch.
A three-layer protocol stack is converging: payment rails (MPP, Coinbase x402), trust and authorization (Google AP2, Visa Trusted Agent Protocol, Mastercard Agent Pay), and commerce (Google UCP, OpenAI ACP). On April 8, Visa launched Intelligent Commerce Connect — a single platform supporting all four competing protocols as a network-agnostic payment layer.
McKinsey estimates AI agents could mediate $3–5 trillion in global transactions by 2030. Juniper Research (April 7) projects $1.5 trillion in agentic commerce spend by 2030. The orchestration layer is the only infrastructure positioned to govern this new class of non-human payers.

The Complexity Tipping Point

Here’s a transaction that happens thousands of times a day and nobody thinks about:

A customer in São Paulo buys software from a SaaS company in Toronto. She pays with a local credit card. The payment routes through a Brazilian acquirer, converts from BRL to CAD, passes a fraud check calibrated for cross-border e-commerce, authenticates via 3D Secure, and settles into the merchant’s Canadian bank account — minus interchange, FX spread, and acquirer fees.

Now multiply that by every payment method the merchant accepts. Pix in Brazil. Interac e-Transfer in Canada. iDEAL in the Netherlands. SEPA Instant across the EU. Apple Pay everywhere. BNPL for the cart abandoners. And the card networks underneath all of it, each with their own routing preferences, decline reason codes, and fee structures.

This is the problem. Not any single transaction — but the accumulation of hundreds of provider-specific integrations, each with its own API, its own settlement mechanics, its own compliance quirks. The average enterprise merchant now integrates 6 to 8 payment methods. Over 81% of global enterprises are running omnichannel strategies that require integration across eight or more methods per platform. Cross-border digital transactions accounted for 36% of all digital transactions in 2023, and B2B cross-border payment value is projected to climb 40% by 2028.

No single PSP can be best at everything, everywhere, for every transaction type. And building direct integrations to each provider — then managing the operational overhead of routing logic, failover cascading, reconciliation across all of them — is engineering quicksand.

Payment orchestration exists because this problem got too big to solve any other way.

What Orchestration Actually Is

Strip away the marketing language and a payment orchestration platform does one thing: it sits between the merchant’s checkout and everything downstream — PSPs, acquirers, gateways, fraud tools, 3DS providers, alternative payment methods — and makes the routing decision.

One API integration. Every provider underneath. The orchestration layer handles the logic: which PSP gets this transaction based on geography, card type, amount, currency, and historical success rates? If the primary provider declines, which backup gets the cascade? Which fraud tool screens it? Which 3DS flow authenticates it? How does it reconcile across providers at end of day?

The capabilities have matured fast. Smart routing engines now use machine learning to optimize in real time — not just rules-based fallbacks, but models trained on millions of transactions that learn which provider-currency-card type combinations yield the highest approval rates at the lowest cost. Adaptive retry logic reads decline reason codes and adjusts cascading strategy accordingly. Portable tokenization vaults store payment credentials PSP-agnostically, so merchants aren’t locked into a single provider’s token format. Fraud orchestration routes transactions through different screening tools based on risk profile — heavy checks for high-risk traffic, lighter touch for known customers.

The performance impact is measurable. AI-based routing modules rolled out in 2024 cut authorization declines by up to 12%, particularly in high-volume verticals like travel and digital subscriptions. Real-time analytics and fraud detection, now deployed by 58% of orchestration platforms, have reduced transaction failure rates by 29% compared to traditional payment setups.

For merchants processing at scale, each percentage point of improved approval rates translates directly into captured revenue that would have otherwise been lost to false declines or suboptimal routing.

Who’s Building It

The competitive map breaks into two camps, and the tension between them is the defining dynamic of the market.

The pure-play orchestrators build the routing layer and nothing else. They’re PSP-agnostic by design — the whole point is that merchants can mix and match providers underneath.

Spreedly is the category pioneer. Founded in 2008 out of Durham, North Carolina, it offers a PCI-compliant orchestration and tokenization platform connecting to over120 payment gateways and PSPs through a single API. Its portable vault — where payment credentials are stored independent of any single provider — is the purest expression of the orchestration thesis: your tokens should be yours, not your PSP’s.

Gr4vy, founded in 2020 by PayPal veteran John Lunn, differentiates on infrastructure architecture. It’s the only orchestration platform that spins up a dedicated cloud instance for each merchant — isolated infrastructure on AWS, GCP, Azure, or private cloud. The pitch is that shared multi-tenant platforms create single points of failure; dedicated instances give merchants full control. Gr4vy has raised $27.2M at a $115M valuation from Nyca Partners (whose bench includes ex-Visa executives), March Capital, and Activant Capital. It recently partnered with Google on an alpha MVP for agentic payments.

Primer, also founded in 2020 and based in London, positions itself as “unified payment infrastructure” — broader than orchestration alone. It connects PSPs, wallets, fraud tools, 3DS, and reconciliation through a single API and offers a no-code workflow builder that lets payment teams design, test, and optimize flows without engineering resources. Strong in travel, gaming, and retail.

Then there are the PSPs building orchestration into their own stacks — which is the competitive threat that keeps pure-play orchestrators up at night.

Nuvei is the Canadian flagship. Founded in 2003 as Pivotal Payments, Nuvei has built a full Payments Orchestration hub on top of its acquiring business. Smart Routing Manager. 3DS Cascading. Fraud and risk management. Chargeback analytics. All accessible through a Control Panel that lets merchants configure routing parameters, manage authentication exemptions, and set custom conversion rules. The platform reaches 200+ markets with local acquiring in 50, supports 150 currencies and 720 alternative payment methods, and handles Canadian-specific rails including Interac Instant. Recent wins include MediaMarktSaturn — Europe’s largest consumer electronics retailer — for marketplace payments, and Nautilus Plus for omnichannel payments across its Quebec gym network.

The Nuvei approach raises the central strategic question in orchestration: do merchants want a neutral routing layer that sits above their PSPs, or do they want their PSP to be the orchestration layer? Adyen’s single-platform model makes the same argument from a different direction — why add a middleware layer when the acquirer can optimize routing internally?

Stripe has taken a third path. Through acquisitions of Bridge (stablecoin orchestration, $1.1B), Privy (wallets), and Metronome (usage-based billing), plus the launch of its Agentic Commerce Suite and co-development of protocols with OpenAI and Paradigm, Stripe is building something closer to the entire financial operating system. Not just orchestration. Not just acquiring. The full stack.

The Twist: Machines Start Paying

Everything above describes orchestration as an optimization layer for human-initiated transactions. Better routing. Higher approval rates. Lower costs. Important, but incremental.

What happened on March 18, 2026 is not incremental.

Stripe and Tempo launched two things simultaneously: the Tempo mainnet — a payments-focused blockchain built for stablecoins and real-world settlement at a $5 billion valuation — and the Machine Payments Protocol (MPP), an open standard that lets AI agents request, authorize, and settle payments programmatically. No human clicking through checkout. No account creation. No pricing page navigation. An agent requests a service, gets the price, pays, and moves on.

Visa extended MPP to card-based payments on its global network the same day, releasing a card-based spec, an SDK for developers, and access to its Trusted Agent Protocol for secure machine payments. Mastercard is building Agent Pay within the same ecosystem and, in the same month, announced a $1.8 billion acquisition of BVNK — a London-based stablecoin infrastructure startup — to embed digital asset rails into its network.

More than 100 services were in the MPP payments directory at launch. Agents can already pay to spin up cloud browsers through Browserbase, print and send physical mail through PostalForm, order food for human pickup at a New York sandwich shop, and contribute to carbon removal projects through Stripe Climate. This is not a whitepaper. It’s live infrastructure.

The protocol race extends well beyond MPP. What’s converging is a three-layer stack:

Payment rails — Stripe/Tempo MPP and Coinbase’s x402 protocol. These are the low-level primitives for moving money between machines. MPP handles fiat and stablecoins. x402 embeds sub-cent USDC payments directly into HTTP requests.

Trust and authorization — Google’s Agent Payments Protocol (AP2), Visa’s Trusted Agent Protocol, and Mastercard’s Agent Pay. These layers don’t move money. They define who an agent is, what it’s allowed to spend, and how that mandate is cryptographically signed and revoked.

Commerce — OpenAI’s Agent Commerce Protocol (ACP) and Google’s Universal Commerce Protocol (UCP, announced by Sundar Pichai at NRF 2026 and co-developed with Shopify, Etsy, Wayfair, Target, and Walmart). These handle product discovery, checkout, merchant-of-record rules, and structured negotiation between buying and selling agents.

These layers don’t compete — they stack. A single transaction might use AP2 mandates to authorize an agent, MPP to settle the payment, and UCP to structure the commerce interaction.

The convergence is accelerating in real time. On April 8, Visa debuted Intelligent Commerce Connect (ICC) — a platform that supports all four competing agentic protocols (Visa’s Trusted Agent Protocol, Stripe’s MPP, OpenAI’s ACP, and Google’s UCP) as a single, network-agnostic payment layer. When an AI agent initiates a purchase through ICC, the platform identifies the card, replaces it with a secure token, verifies the agent is operating within the consumer’s spending constraints, and routes the transaction to the appropriate network. Any major credit card works, not just Visa. AWS is among the first pilot partners. Separately, Mastercard has expanded its Agent Pay capabilities to Hong Kong as part of a broader push to build an international agentic commerce network.

Visa positioning itself as the neutral payment layer underneath the protocol competition — ensuring that no matter which standard wins, agentic transactions still flow through its network — is a strategic tell. The card networks aren’t waiting to see how agentic commerce plays out. They’re building the plumbing now.

The scale projections are hard to ignore. McKinsey estimates agent-mediated purchasing could account for $3–5 trillion in global GMV by 2030. Juniper Research, in a report published April 7, projects $1.5 trillion in agentic commerce spend by 2030 — while noting that trust remains the number one barrier to deployment. McKinsey projects up to $1 trillion in US retail revenue alone. Salesforce data shows AI-driven interactions already influenced roughly $67 billion in global online sales during Cyber Week 2025 — approximately 20% of all global Cuber Week orders.

Why Orchestration Becomes Non-Negotiable

Here’s the part that the pure market-sizing misses.

Every assumption in today’s payment stack — that a human clicks “buy,” that a browser renders a checkout page, that strong customer authentication involves a person — breaks when the payer is software. AI agents generate transaction patterns fundamentally different from human browsing: higher speed, concurrency, no mouse movements, no session cookies. They need programmatic authorization, policy-driven routing, and machine-readable payment interfaces.

The orchestration layer is the only infrastructure positioned to handle this:

Policy enforcement. Agents operate within user-defined constraints — budget caps, whitelisted merchants, category restrictions, time windows. The orchestration layer evaluates these policies in real time and routes or blocks accordingly. This isn’t optional. Without it, you have uncontrolled machine spending.

Dynamic rail selection. An agent buying on behalf of a user might choose between card rails, stablecoin settlement, real-time payments, or BNPL depending on cost, speed, and merchant acceptance. Only an orchestration layer can make this selection dynamically per transaction.

Fraud differentiation. Agent traffic needs different fraud models than human traffic. Orchestration platforms can route agent-initiated transactions through specialized screening workflows — heavier checks for unfamiliar agents, lighter touch for verified ones operating within known parameters.

Trust verification. Every agentic transaction requires verification of who acted, under what policy, and within what limits. The orchestration layer mediates this trust — and creates a new revenue stream. Even micro-fees for identity verification and policy evaluation, priced at fractions of a percent, translate into hundreds of millions at agent-commerce scale.

Everest Group puts it cleanly: the control plane of payments is shifting upstream, from “click to pay” to “configure and let act.” Whoever owns the orchestration layer owns the payment control plane of the agent economy.

The Regulatory Catch

The main barrier to fully autonomous agentic payments is not technological. It’s regulatory.

PSD2 and Strong Customer Authentication in the EU and UK require clear human authorization for payment orders. There’s no current mechanism that treats an AI agent as equivalent to a human payer. The EU AI Act, with major obligations taking force in 2026, requires risk classification, human oversight, accountability, and transparency for AI systems — but doesn’t yet permit fully autonomous payments.

Industry bodies including UK Finance have raised several unresolved questions: who is ultimately liable when an AI system makes autonomous decisions? How is meaningful human oversight maintained in multi-agent systems? How is responsibility shared across third-party AI supply chains?

The practical result is that most payment companies in 2026 are running a “human-in-the-loop” approach. AI prepares the purchase. The user approves the payment. Truly autonomous agent spending operates within strict, user-defined guardrails. Visa’s Trusted Agent Protocol and Stripe’s Shared Payment Tokens are designed precisely for this transitional architecture — scoped, time-limited, context-bound authority.

This will be a long transition. But the infrastructure is being built now, and the orchestration layer is where the governance lives.

The Stakes

Payment orchestration started as middleware. A convenience layer for merchants tired of managing multiple PSP integrations. The market sizing tells one story — $2–10 billion by the early 2030s, healthy growth, good venture returns.

The strategic significance tells a different story entirely.

In a world where AI agents mediate trillions in transactions, the orchestration layer isn’t middleware. It’s the control plane. The layer that determines which rail settles a transaction, which fraud model screens it, which policy governs the agent initiating it, and whether the transaction happens at all.

Stripe is betting on this with everything — acquisitions, protocols, a purpose-built blockchain. Visa and Mastercard are extending their trust networks into agent identity. Google is building the commerce layer. The pure-play orchestrators — Spreedly, Gr4vy, Primer — are racing to prove they can govern agent transactions as effectively as they route human ones.

The question isn’t whether payment orchestration matters. The question is who controls the transaction when the buyer isn’t human.

That’s the $3 trillion routing decision.

Sources

Turbostream

Apr 12

The orchestration layer being the governance point for autonomous agent payments is a really important insight. x402 solves the "how do machines pay" question at the protocol level, but orchestration solves the "which payment method, which route, under what conditions" question, which is where the real business logic lives.

For AI agents operating across multiple chains and payment rails simultaneously, a sophisticated orchestration layer becomes the trust and policy enforcement point that the base protocol lacks. This actually suggests payment orchestration providers have a strong strategic position in the agentic economy, potentially more durable than being a pure payment processor, since they sit between the intent and the execution in a way that creates natural leverage.

Fintech Observer

Discussion about this post

Ready for more?