By Shona Sabah (Senior Manager - Strategic Growth Lead)
At the heart of every payment professional’s challenge lies a question that’s deceptive in its simplicity, yet finding the answer has proven to be tricky...
"How do we make transactions more successful, lower-cost, and lower-friction, all while expanding global reach and reducing risk?"
For years, increasingly innovative orchestration has been our recurring response to this conundrum. We’ve layered processors, built smarter routing, added retries, and supported local payment methods. It’s helped make progress, there’s no denying that, but these systems, which have delivered success for so long, all depend on humans writing the rules.
In the face of a new wave of transformation powered by agentic AI (autonomous, self-learning systems capable of making and executing decisions on behalf of users, merchants, and even other systems), that must now change.
Whereas AI in payments has traditionally been about analysis and recommendation, agentic AI is about acting. For example, rather than flagging that a card failed and prompting a human to retry, agentic AI can autonomously route the transaction differently, switch acquirers, retry at a more optimal time, or even onboard a new local payment method, all in real time and with minimal human input. In simple terms, AI in payments no longer just learns, it ‘does’ and that shift from ‘insight’ to ‘initiative’ is having, and will continue to have, profound implications for how payments are orchestrated.
Over the next one to three years, we’ll see early examples of agentic AI embedded directly into orchestration platforms. For PSPs and payment platform providers, this will unlock new ways to solve persistent pain points and create new forms of differentiation. It’s far from an exhaustive list, but I thought it would be useful to explore just a few of these:
1. Tackling declines and false negatives
Even the best orchestration stacks out there still lose good transactions, either through soft declines, issuer quirks, network blips etc. The research shows that good orchestration can already cut false declines by 10–15% for UK enterprises, which is great, but agentic systems can take that further by continuously monitoring performance, detecting anomalies, and adjusting routing as they happen, all without a human ever touching a rule dashboard. This means higher acceptance rates, fewer failed checkouts, and less manual firefighting.
2. Simplifying operational complexity
As merchants expand into new geographies, payment types, and wallets, their orchestration logic often becomes unmanageable. The rules multiply and exceptions grow, meaning performance tuning becomes a full-time job for somebody. Agentic AI offers what we might call a ‘control tower’ model; one that learns patterns, optimises flows, and manages exceptions dynamically. So, instead of humans writing rules, teams set objectives (e.g. maximise success rate, minimise cost, balance across providers) and the agent does the rest.
3. Reducing payment leakage and involuntary churn
We all know that merchant revenue can quietly leak away when failed payments go unaddressed. Current orchestration stacks certainly go some way to solving this issue - one orchestration-vendor cited a case where failure-routing uplifted recovery from 51% to 66% in one month - but again, Agentic AI takes that a step (or several steps!) further by spotting ‘at-risk’ renewals, retry at the most likely time to succeed, or even prompt users to update payment credentials before revenue is lost. And, you guessed it, it’s all autonomous.
4. Enabling New Commerce Models
It’s already clear that we’re entering an era where it won’t necessarily be humans who are initiating every payment. Conversational interfaces, embedded commerce, and AI assistants are creating agent-to-agent transactions that are growing in usage and popularity. The examples that spring to mind for me are a travel assistant that books and pays for your flights automatically, or a B2B procurement bot that negotiates and pays suppliers based on live cashflow data. Agentic AI is the underlying infrastructure that makes those flows possible and it’s creating a world where agents transact securely, with consent and traceability built in.
For Merchants
For merchants, the value of agentic AI lies in its ability to directly improve performance and efficiency. By dynamically adjusting routing, retry logic, and acquirer selection, agentic systems can lift conversion rates in measurable ways. One provider, for example, reported an average 3.8% increase in acceptance "simply" through AI-driven routing. Those small percentage gains translate into millions in recovered revenue at scale.
Beyond conversion, agentic AI helps tackle involuntary churn, something that’s been a constant headache for many businesses for some time. When payments fail, agents can automatically identify the cause, retry at the optimal time, or prompt customers to update credentials, all without manual intervention. That means fewer failed payments and less revenue leakage.
For global merchants, scalability becomes easier too. Agentic systems learn from live data, continuously optimising which acquirers, methods, or payment routes perform best in each market. This allows international expansion without the burden of manually tuning rules or managing complex exceptions.
And finally, lower operational costs. By reducing the manual overhead associated with monitoring dashboards, maintaining rule libraries, or reacting to outages, agentic orchestration allows teams to focus their time on higher-value work like strategy, partnerships, and customer experience.
For Consumers
Consumers may never see the term agentic AI, but they’ll feel its impact in the form of smoother, more reliable experiences. Fewer failed checkouts, fewer payment interruptions, and fewer moments of friction all add up to stronger trust and higher satisfaction.
As commerce increasingly shifts to voice, chat, and assistant-driven interactions, agentic AI will quietly power those experiences. Imagine asking your digital assistant to renew a subscription or pay a bill, and it completes the entire process, securely and accurately, without redirecting you or asking for another authentication step. That’s the kind of invisible convenience consumers will come to expect, and agentic AI enables it.
Despite the technological advances, security and trust will remain critical. The best implementations will acknowledge that and bake in consent, credentials, and traceability from the start. Mastercard’s Agent Pay framework is an early example as agents are registered, tokenised, and their actions can be audited. Done well, agentic AI can actually enhance consumer trust by making payment flows simpler and safer.
For Payment Providers
The first thing to say here is that, for payment providers, the emergence of agentic AI is a commercial opportunity that cannot be ignored. Those who act fast and embed self-optimising orchestration into their platforms will gain a real source of differentiation as merchants will increasingly choose providers who can deliver higher acceptance, lower cost, and less operational drag automatically.
The second point to flag is that the sooner you act, the bigger advantage you’ll continue to have in the future. That’s because every transaction that’s processed through these systems will feed a powerful data flywheel. Over time, your models get smarter, the routing gets more efficient, and the performance gap between agentic and rule-based systems widens. That creates a compounding network advantage i.e. the more volume you handle, the better your engine becomes.
Obviously, the financial benefits are clear too. Smarter routing reduces acquiring and interchange costs, and fewer failed transactions mean fewer retrials, chargebacks, and reconciliation efforts. All of this directly improves your margin and reduces cost-to-serve.
I've said it already, but I'll say it again; early adopters will be better positioned for the inevitable coming shift toward agentic commerce and a world where not every payment is initiated by a human. By investing now, PSPs can future-proof their technology and product roadmap for an environment that will very soon look very different.
As with any major shift, the road to agentic orchestration comes with technical, organisational, and ethical hurdles which you’ll need to overcome.
The first is data quality and model trustworthiness. Agentic systems rely on vast amounts of high-quality data to make accurate, real-time decisions, yet we all know that payments data is often fragmented across acquirers, issuers, and geographies. Any incomplete or biased data can lead to poor outcomes, or even systemic errors. Building trustworthy models means investing in clean data pipelines, continuous monitoring, and human-in-the-loop oversight.
Complex risk, fraud, and compliance challenges also regularly emerge in an agentic world. For example, when an autonomous agent makes a payment or reroutes a transaction, who is accountable if something goes wrong? PSPs will need to design robust governance frameworks that track intent, authorisation, and action to ensure every agent is properly credentialed and auditable before handing over decision-rights. Mastercard’s token-based approach to agent credentials is a promising start, but it’s only one part of a much larger puzzle.
Then there’s legacy technology and integration complexity Most existing orchestration platforms were built on rule-based architectures, not real-time decisioning engines. Retrofitting those systems to accommodate agentic AI will likely be complicated and costly. Providers will need to modernise their routing logic, integrate richer telemetry, and build decision layers that can act autonomously without breaking downstream systems.
The merchant mindset presents another challenge. For many, handing over control to an AI agent (even a well-trained one) will feel uncomfortable. Moving from ‘we define the rules’ to ‘the system learns and acts’ requires large amounts of trust and transparency and a new operational philosophy. Providers will need to design clear interfaces and controls that help merchants understand why the agent made a given decision and how to intervene if needed.
As agentic payments mature, competing frameworks and protocols are already emerging, so ecosystem standardisation must also come into play. Without interoperability, the industry risks fragmentation, driving up cost and slowing adoption. Shared standards will be essential to ensure agents can transact safely across networks and geographies.
Finally, and perhaps most importantly, consumer trust will be the deciding factor. For many users, the idea of an AI making payments on their behalf will require reassurance and transparency about what agents can do, explicit consent for what they control, and simple opt-out mechanisms will be vital. If trust is lost early, meaningful adoption could stall for several years.
If you’re leading strategy, product, or marketing in a payment provider, this shift raises three critical questions.
How will your orchestration evolve from rules-based to agent-driven?
What new value propositions could this unlock, and how can we monetise them?
What governance needs to be in place?
Over the next few years, I expect agentic AI to rapidly move from the initial experiments we’re seeing today, to being fully embedded across the payment's ecosystem. Payment providers who act early (yes, there is still time) and embed autonomous decision-engines will help their merchants unlock hidden revenue, reduce operational costs and deliver smoother consumer experiences. Of course it won’t be easy; they will also need to manage new risks, build new capabilities and re-shape their roadmap, but that’s the price you’ll need to pay if you want to lead the charge into this brave new world.