The AI investment paradox that every payments leader needs to confront
Organisations are spending more on artificial intelligence than at any point in history. At the same time, industry surveys consistently show that large-scale AI projects are failing to deliver the returns that were promised. In the payments industry, where the entire business model is built on keeping volume, reducing unit cost, and increasing revenue in careful balance, the gap between investment and outcome is not just a strategic embarrassment. It is an existential risk.
I explored this tension in a recent conversation on the AI Adoption Podcast with Rob Lunn, a payments professional with experience across large international banks, early-stage ventures, and now Finality, a regulated wholesale payment system that settles transactions in central bank money using blockchain technology. The conversation was one of the most practically useful I have had in this series, because Rob did not deal in abstractions. He dealt in mechanics.
The two technologies are not interchangeable
One of the most persistent errors I encounter in boardroom discussions about technology adoption is the assumption that AI and blockchain are somehow competing tools, or that choosing one means deprioritising the other. Rob's framework dismantles this immediately.
Blockchain, he argues, delivers its greatest value at the moment of settlement. Settlement risk, the risk that one party in a transaction fails to deliver while the other has already paid, is one of the most costly and systemically dangerous problems in wholesale financial markets. The collapse of Lehman Brothers demonstrated what happens when that trust breaks down at scale: liquidity freezes, counterparties stop transacting, and the damage spreads far beyond the original failure.
Blockchain addresses this through atomic settlement. In plain terms, atomic settlement means that payment and asset transfer are legally and technically linked as a single event. One cannot occur without the other. A bank paying sterling for dollars cannot lose the sterling without receiving the dollars, because the transaction is structured so that both sides complete simultaneously or neither completes at all. The distributed ledger, a shared record visible to all participants simultaneously and updated by consensus rather than by a central authority, provides the certainty and trust that makes this possible.
AI, by contrast, operates in the space of uncertainty. It is well suited to reading unstructured data from invoices, aggregating vendor records, forecasting liquidity positions, and helping treasury teams understand not just what they owe, but when and with what they should pay it. It earns its value at the beginning and end of the payment process, where data is rich but insight is poor. At the settlement point itself, where absolute certainty is required, AI has relatively little to contribute.
For leaders, the practical implication is clear. These technologies have different jobs. Conflating them, or deploying AI at the settlement layer because it feels more modern, is a category error with real cost consequences.
The paradox that should worry every board
The AI investment landscape is characterised by a paradox that Rob identified with unusual directness. Investment appetite is high. Success rates are low. Events such as London Tech Week and the broader body of industry surveys confirm both sides of this equation simultaneously.
In the payments industry, this paradox is particularly dangerous. Payments is, at its core, a scale processing business. The unit economics are simple: increase volumes, reduce unit costs, protect or expand margin. Every significant deployment decision needs to be tested against that logic. An AI project that consumes capital, fails to deliver efficiency, and erodes customer confidence does not just waste money. It shifts the unit economics in the wrong direction, which affects pricing, competitiveness, and ultimately volume.
The discipline Rob advocates is not a reduction in ambition. It is a sharpening of it. A small number of well-defined projects. Clear criteria for success. A rigorous assessment of total cost of ownership that includes the customer outcome, the regulatory obligation, and the long-term capability being built. Organisations that treat AI as a broad infrastructure investment, deploying it widely in the hope that benefits will follow, are the ones most likely to find themselves in the statistics on project failure.
Alexandra Dobra Kiel https://open.spotify.com/episode/7ppGRmhvInuKFoCyiVscOW?si=jFQAB9cqT1urfe12njbGRg
The talent pipeline is the underestimated risk
Of everything Rob raised in our conversation, the talent question is the one I think receives the least attention relative to its importance.
There is a reasonable concern in many organisations that AI will displace entry-level roles. Rob does not dismiss this. He confirms it, and extends it beyond the back office. Front-office functions, including sales processes, request-for-proposal responses, and account management documentation, are equally exposed to automation by generative AI. The entry points into the payments industry are changing across the board.
The problem is not the automation itself. Automation in payments is not new; card acquirers and app stores have been running near-fully automated payment processes for years. The problem is what organisations are planning to do about the talent pipeline that would otherwise run through those roles.
The payments leaders of the future will need to be technically literate without being software engineers. They will need to understand the digital asset space, the principles of distributed ledger technology, the capabilities and limitations of AI, and the mechanics of the risks that sit inside payment systems. They will also need deep domain expertise in payments: understanding settlement risk, managing counterparty relationships, and operating within a demanding regulatory environment.
Organisations that remove entry-level roles without investing in the alternative pathways for developing that knowledge are borrowing against a capability debt that will eventually come due. The middle and long-term consequences, a management layer that lacks domain depth, that deploys AI without understanding its limits, that makes decisions that inadvertently increase regulatory risk, are serious.
The regulatory relationship is a strategic asset
One area where payments organisations have an advantage over other sectors is in their established relationship with regulators. Rob's approach at Finality is instructive: engage regulators early, explain the proposition clearly, and demonstrate a positive contribution to the ecosystem the regulator is responsible for.
This is not simply risk management. Regulators are increasingly aware that their approach to oversight must allow innovative technologies to deliver benefits, not merely constrain them. The FCA's use of regulatory sandboxes reflects exactly this awareness. Organisations that invest in that relationship, that help regulators understand the business model and the systemic benefit, are better positioned to move quickly and confidently when the regulatory environment evolves.
For leaders, this means the compliance team is not a gatekeeper to be managed around. It is a strategic partner in building the case for responsible innovation.
The question I am leaving with
The payments industry is a useful lens for a much larger question. If a business is built on a logic as clear as volume, unit cost, and margin, and organisations are still struggling to deploy AI effectively within it, what does that suggest about sectors where the underlying logic is harder to articulate?
The organisations that will benefit most from AI and blockchain are not those with the largest technology budgets. They are the ones with the clearest sense of the problem they are solving, the discipline to match the tool to the task, and the foresight to protect the talent pipeline that will be needed to manage these systems responsibly in ten years' time.
I would welcome your reflections. The conversation with Rob reinforced my view that the most important AI decisions are not technical. They are strategic, organisational, and, ultimately, human.
Listen to the full conversation with Rob Lunn on
Comments